The Information-Asymmetry Week: Researching Employers Properly
WEEK 103 :: POST 3 :: CLAUDE
Directions Given To The A.I. This Week+
Instructions Given to each A.I. — Please provide 3 prompt variations that share this objective:
Each A.I. also received two static attachments: the blog post template (structure) and the authoring instructions (voice and standards). The text below is the week-specific assignment as sent — reflowed for the web; wording unchanged.
I'd like you to write this week's Ketelsen.ai post. Two files are attached: the blog post template (the structure to follow) and the authoring instructions (context, voice, and standards). Please read both before you begin, then produce the complete post in a single response.
This week's theme: "Company Intelligence — Researching Employers Like a Pro."
This is Week 4 of an eight-week series on running a job search with AI, and it is the series' information-asymmetry showcase — the job-search version of the car series' dealer-research week. An employer entering an interview knows the reader's history, salary expectations, and references. The reader knows the employer's careers-page marketing. This week closes that gap: building a ranked target-company list, and screening out bad employers before investing application effort in them.
The territory is rich because the failure modes are real and documented: job listings that stay up with no one hiring behind them (the "ghost listing" — a direct parallel to the ghost cars of the last series), suspiciously coordinated bursts of five-star employer reviews, serial layoff patterns, and the culture warning signs that only show up when you know to look — leadership churn, vague answers about why the role is open, review complaints that cluster on the same theme across years.
The deliverable the reader should walk away holding: a ranked target-company dossier — this series' "Dealer Intelligence Matrix" moment — with each company scored against the Week 2 spec and screened for red flags.
THE SERIES CONTRACT — identical every week; it binds every prompt you design. This series' tagline is its editorial contract: "Use AI like an analyst, not a ghostwriter." It is written for a reader in a market unsettled by AI itself — some readers are searching precisely because AI eliminated their last role. Write with that reader at the table: no AI-efficiency cheerleading, no automation jokes, no promises that AI will "do it for you" anywhere a human hiring decision is involved. And hold one line in every prompt: the AI is the reader's private analyst, coach, and sparring partner — it structures, researches, rehearses, and questions. It does not ghostwrite. Anything a hiring human will read or hear — résumé lines, cover letters, outreach messages, interview answers, negotiation emails, resignation letters — must end in the reader's own words and be true. Prompts should drive toward drafts the reader rewrites and owns, and should say so explicitly. Employers increasingly restrict how AI may be used in their own hiring decisions for legal and compliance reasons, and recruiters increasingly recognize — and discard — material that reads machine-written. A prompt that makes a reader look AI-generated hurts them twice. Posts that ignore this contract should expect to lose the week. Two practical notes. First: a standing “About this series” notice covering these same points is added to every published post automatically at publication — acknowledge the frame in your own voice where your week's prompt calls for it, but do not write a formal disclaimer block of your own, and do not open every post with the same acknowledgment paragraph: outside the weeks whose prompts explicitly carry the series frame, this contract lives in your tone and your prompt design. Second, the framing is POSITIVE: used this way — analyst backstage, reader on the page — AI is an advantage no hiring human will ever hold against your reader. Write like that is true, because it is.
The three prompts should help a reader:
- Build the company snapshot brief. A repeatable research checklist for one company: what it does, how it makes money, how it is doing, what current and former employees say — organized as questions with named places to look (filings and annual reports for public companies, funding announcements for startups, employer-review sites read with dates in mind), so the reader assembles the evidence.
- Run the red-flag audit. A structured pass over one company for warning signs: layoff history, leadership turnover, funding runway for startups, and review forensics that weight recency and pattern over any single angry post — each flag framed as something to verify, with the audit distinguishing "confirmed," "concerning," and "rumor."
- Build the full dossier matrix. The multi-company system: a research workflow the reader runs per company, feeding a matrix that scores each against the Week 2 must-haves and deal-breakers plus the red-flag audit, producing a ranked list of where to actually spend application effort — and a screened-out list with reasons.
At the advanced tier, the strongest version of this week is the dossier as a living system — a research routine organized around the areas that matter (financial health, review patterns, listing legitimacy, pay transparency, leadership signals, growth trajectory, interview-process reputation, and which companies hire by referral before ever posting), feeding one composite-scored matrix. That architecture is worth reaching for; it is the week readers will bookmark.
The constraint that defines this week. AI models cannot see live listings, current filings, this week's layoffs, or yesterday's reviews — and several posts in the last series were marked down for letting the AI assert facts it could not know. No prompt may ask the AI to state a specific company's current facts — layoffs, funding, leadership changes, whether a posting is real — as settled fact. The prompts should have the AI build the research structure: the questions, the named source types, the scoring framework, the red-flag taxonomy. The reader gathers the evidence; the AI organizes it and helps score it. A prompt that has the AI "investigate" a named company and report findings should expect to be marked down on Content Accuracy — the honest version has the reader paste in what they found.
Series dependency chain, for the Metadata block: Week 4 consumes Week 2's target-role spec sheet (companies are scored against its must-haves and deal-breakers). Week 4 produces the ranked target-company dossier — consumed by Week 5 (applications go to the top of the ranking), Week 6 (interview prep draws on the dossier), and Week 7 (the negotiation leans on what the reader knows about the company's position).
Because readers may arrive at this post without having read the earlier weeks, the prompts should work for someone with a shortlist scribbled on a napkin, while making clear the matrix bites harder when it scores against a real Week 2 spec.
Three difficulty tiers as always — Beginner, Intermediate, Advanced — each a genuinely different approach to the same problem, not the same prompt at three lengths.
On examples: this is a career topic that touches every industry. The template lists tech startup / retail / freelance as suggested industry examples — those are marked MAY, and adapting them is expected here. A reader vetting a fast-growing startup's runway, another weighing a big company's layoff pattern against its brand, and a third trying to tell a real posting from a ghost are the right kinds of contexts. Choosing them over the suggested business examples is correct behaviour and will not be scored against you.
A note on supplied figures. Anything marked `[SUPPLIED — use as given]` above came from Ketelsen.ai's own research brief. Use it freely — you are not fabricating by repeating it, and you will not be marked down for leaving it uncited. Do not attach an invented source to it. (No supplied figures this week — and given the constraint above, this is the worst possible week to invent one. Ghost-listing prevalence numbers circulate widely and vary wildly; write around them.)
## BEFORE YOU SUBMIT — STRUCTURAL CHECK
(This block is identical every week. It exists because these specific items are the ones posts drop, and a dropped structural item costs compliance points for something that takes one minute to add.)
Your post is parsed by a script before any human reads it. Confirm all seven:
1. ☐ Response begins with `PLATFORM: <your name>` and `WEEK: 4` 2. ☐ `## Lead` present once, at the very top, before Variation 1 3. ☐ `## In one line` present in all three variations 4. ☐ `## What this prompt gives you` present in all three variations 5. ☐ `## The Prompt` present in all three variations, with the prompt in double quotes beneath it 6. ☐ `## Introductory Hook` and `## Current Use` present in all three variations (three of each — not one) 7. ☐ Every template heading written as `##`, none bolded instead; prompt breakdown is running text split on ` : `, with no `###` headings inside it
A complete post has 57 `##` headings. If your count is well short, a section is missing or was bolded instead of hashed.
One extra check this week: confirm no prompt asks the AI to report a named company's current facts — layoffs, funding, leadership, listing legitimacy — as findings. The AI builds the research structure and scores evidence the reader gathered; the reader does the investigating.
About this series. AI is reshaping work — for some readers, it's part of why you're searching at all. We don't pretend otherwise. And the hiring world is wary of AI-written material: many employers restrict how AI may be used in their own hiring decisions, and recruiters increasingly recognize — and discard — machine-written applications. So this series teaches a different approach: use AI as your private research analyst, interview coach, and thinking partner, while every word an employer sees or hears from you stays genuinely, verifiably yours. AI behind the scenes. You on the page.
The employer walks into the interview holding your résumé, your salary history, and three references. You walk in holding their careers page. This week closes that gap with three prompts that turn employer research into a repeatable system: a five-part company snapshot brief you can finish in an afternoon, a red-flag audit that sorts what you found into confirmed, concerning, and rumor, and a scored dossier matrix that ranks every company on your list and tells you which ones to stop wasting applications on. None of them ask the AI to tell you facts about a company — that is the whole trick, and the reason they work.
The Company Snapshot Brief
Turn a careers page into a research plan you can finish.
Most people research an employer the same way. They read the About page, skim the careers page, glance at a review site, and decide the company seems fine. That is not research. That is absorbing marketing written by people whose job is to make you feel this way, plus a handful of anonymous reviews with no dates attached. Meanwhile the company's recruiter has your work history, your references, and a compensation band they will not show you. The asymmetry is not close, and it does not correct itself during the interview — it gets worse, because by then you are invested and they are not. The fix is unglamorous. You need a checklist that tells you what to find out and where to look, and then you need to actually go look.
Why this matters now
The application itself is the cheap part now. Sending twenty applications takes an afternoon, which means the scarce resource in a job search is no longer effort — it is judgment about where to point the effort. Every hour spent on a company that was never going to work is an hour not spent on one that would. This prompt costs you five minutes and produces the structure for a research session you can run before you write a single line of an application. It also has a second payoff people underestimate. The reader who has done this research asks different questions in the interview, and interviewers notice immediately.
You are helping me research a company I am considering applying to. I will give you the company name and the little I already know.
Important instruction, please follow it exactly. Do not tell me facts about this company. You may not have current or accurate information about it, and a wrong fact here would cost me real time and possibly a real opportunity. Your job is to build the research checklist. My job is to fill it in.
Company: [COMPANY NAME]
What I already know: [ONE OR TWO SENTENCES — what they seem to do, where I saw the job, anything a friend told me]
Role I would be applying for: [JOB TITLE]
Company type, if I know it: [PUBLIC / PRIVATE / STARTUP / NONPROFIT / GOVERNMENT / NOT SURE]
Produce a one-page Company Snapshot Brief with these five sections. Write each section as questions for me to answer, not as answers.
1. What they do — the business in plain language, and who their customers are
2. How they make money — the revenue model, and who actually writes the checks
3. How they are doing — signals of health or trouble
4. What it is like to work there — what current and former employees say
5. Why this role is open — what I should try to find out, and what I should ask
For every question, name the specific type of source that answers it. Be concrete about source types — annual and quarterly filings for public companies, funding announcements and investor pages for startups, tax filings for nonprofits, employer review sites read with the posting dates in mind, local business press, the company's own engineering or product blog, the profiles of people who currently hold the job I want. For each question, tell me what a good answer looks like and what a worrying answer looks like, so I can recognize either when I see it.
Then finish with two things. First, three interview questions I could only ask well if I had done this research. Second, a note on which of the five sections I should do first if I only have twenty minutes, and why that one.
How the AI reads this prompt
Practical examples from different industries
The scenarios below are illustrative, written to show how the prompt behaves in different situations. They are not accounts of real companies.
Someone considering a fast-growing startup.
A designer is weighing a role at a forty-person startup that raised a Series A eighteen months ago. She fills in the company name, notes that she found the listing through a friend, and marks the type as startup. The brief she gets back sends her toward funding announcements and the investor's portfolio page for section three, and toward the founders' own public writing for section four. The most useful line in the whole output turns out to be under "why this role is open" — a question about whether the role is new or backfilled, and a note that a backfill at a company this size is worth asking about directly. She asks. The answer is honest and reassuring, and the fact that she asked at all changes the tone of the conversation.
Someone weighing a large employer's stability.
A supply chain analyst is looking at a role at a public company that has been in the news for cost-cutting. He marks the type as public. The brief points him at the annual report's risk factors section and at the discussion where management explains its own results, and it tells him what a worrying answer looks like — the same risk described in escalating language across consecutive years, or a segment whose revenue is described in every filing without ever being given a number. He does not need to become a financial analyst. He needs to read two documents with three questions in hand, which takes an hour and tells him more than a month of guessing.
Someone applying to a nonprofit.
A program manager is applying to a mid-sized nonprofit and realizes she has no idea how to evaluate one. The brief adapts to the type she selected. It sends her to the organization's public tax filings for the money questions, to its annual report to donors for the mission questions, and to the tenure patterns of current program staff for the culture questions. Her most valuable finding is not a red flag at all — it is the discovery that the organization's funding is concentrated in two grants, which becomes an intelligent question about the role's runway rather than a reason to walk away.
Creative use case ideas
Researching the client, not the employer.
Freelancers and contractors have the same asymmetry problem with worse consequences, because a bad client can end in unpaid invoices rather than a bad quarter. The same five sections work with "why is this project open" swapped in for "why is this role open."
Preparing for the reverse-interview half.
Run the brief the night before an interview and mine the questions you could not answer from public sources. Those are exactly the questions to ask when they turn it over to you, and they are more impressive than anything from a list of good interview questions.
Deciding whether to stay.
Point the brief at your current employer. Most people know their team well and their company barely at all, and running the same research on a place you already work is uncomfortable in a way that is usually informative.
Vetting an organization before you volunteer or join a board.
Non-business use. People give community organizations, school boards, and volunteer groups years of unpaid time with less diligence than they would apply to buying a used bike. The five sections translate cleanly — what they do, how they are funded, how they are doing, what people who left say, and why this seat is open.
Choosing a graduate program or trade school.
Swap "employer" for "program" and the structure holds. Completion rates, where graduates actually end up, faculty turnover, and how the program is funded are the same five questions wearing different clothes.
Adaptability tips
Change the five sections to match what you actually care about. A reader who has been laid off twice might replace "what they do" with "who owns them and what happens if that changes," because ownership structure is where their real risk lives. The prompt does not care which five you pick.
Adjust the depth by adding a time budget to the prompt. Add a line saying you have ninety minutes total and want the brief scoped to fit, and the output tightens considerably. Models respond well to explicit budgets and poorly to implied ones.
Run it in batch. Paste in three company names at once and ask for the shared checklist plus a short note on what differs for each, which is faster than three separate sessions when the companies are in the same industry.
For international employers, add a line naming the country. Disclosure regimes differ enormously — the filings that exist for a US public company have different names, different contents, and different timing elsewhere, and a model told the jurisdiction will name the right documents.
Pro tips
Ask the model to mark each question with how confident you should be in the answer once you find it. A number from an annual report and a claim from an anonymous review are not the same kind of evidence, and labeling them at the point of collection saves you from weighting them equally later.
Add one line at the end asking what this checklist is systematically likely to miss. The answers tend to be worth reading — culture at the specific team level, the actual manager, and anything the company has never had a reason to disclose.
Keep your filled-in briefs. By the fifth one you will notice which sections you always skip, and that is either a section you should delete or a blind spot you should fix.
Prerequisites
You need a company name and a job title. That is genuinely all. If you have a target-role spec sheet from Week 2, keep it beside you — the brief is more useful when you already know what you are looking for. You will also need somewhere to write down what you find, and a plain document beats a clever tool here.
Required tools
Any general-purpose AI chat tool on its free tier. No web browsing, plugins, or paid features required — the prompt deliberately does not ask the AI to look anything up. You will need a browser for the research itself.
Frequently asked questions
Why not just ask the AI to research the company directly, especially if my tool can browse the web?
You can, and with a browsing tool you will get something real. But there are two problems worth knowing about. First, a browsing model summarizes what it found in one pass, and the judgment about which sources deserve weight is exactly the judgment you are trying to develop. Second, the summary arrives with no dates attached to any of it, and in employer research dates are most of the value. Running the checklist yourself is slower and produces a better picture. Use browsing for the lookups, and keep the interpretation.
What if the checklist asks for something I cannot find?
That is a finding. A public company whose segment revenue you cannot locate is telling you something; a startup with no public evidence of customers is telling you something louder. Write "not found" rather than skipping the line, because the pattern of gaps across several companies is itself informative. If a whole section comes up empty, ask the model for three alternative source types for that section specifically.
The output feels generic for a small local company. What do I do?
Add detail to the "what I already know" field and re-run it. Small private employers leave almost no paper trail, so the checklist has to redirect toward different sources — local press, the profiles of people who work there now, customers who mention them publicly, and, most effectively, a conversation with someone who worked there. Tell the model the company is small, local, and private, and ask it to build the brief for a company that will not appear in any database.
How long should this actually take?
Plan for sixty to ninety minutes per company on the first few, dropping toward thirty once you know your way around the sources. If it is taking three hours you are reading rather than researching. The brief is a set of questions with answers, not a report.
Recommended follow-up prompts
The red-flag audit
— Variation 2 of this post. Run the snapshot brief first, then feed what you found into the audit to find out which parts of it should worry you.
The interview question generator
— take the gaps in your completed brief and turn them into questions for the interview. The things you could not find out from public sources are the highest-value questions you can ask.
The Week 2 spec sheet check
— score your completed brief against your target-role must-haves and deal-breakers before you write the application, not after.
Tags and categories
Tags:
company research, job search, employer due diligence, research checklist, interview preparation, beginner prompts, information asymmetry Categories: Career and Job Search, Research and Analysis
Citations
U.S. Securities and Exchange Commission, Office of Investor Education and Advocacy, "How to Read a 10-K/10-Q" — Investor.gov investor bulletin. Explains what a public company's annual and quarterly reports contain and which sections describe the business, its risks, and management's own view of results.
U.S. Securities and Exchange Commission, EDGAR full-text search — sec.gov/edgar/search. Free public database of filings by US public companies.
Internal Revenue Service Form 990, as published through ProPublica's Nonprofit Explorer — projects.propublica.org/nonprofits. Public financial filings for US nonprofit organizations.
The Red-Flag Audit
Sort what you found into confirmed, concerning, and rumor.
Research produces a mess. You end up with eleven browser tabs, four review excerpts, a headline you half remember, something a friend of a friend said, and a job listing you are fairly sure has been up since spring. Somewhere in that pile is a real signal, and somewhere else in it is one furious person having the worst week of their professional life. Most people resolve this by mood — the pile feels bad, so they pass, or the salary is good, so they discount the pile. Neither is a decision. The discipline you need is the one investigators use, which is not about finding more evidence but about grading the evidence you already have. Some of it is documented. Some of it is a pattern. Some of it is a stranger's grudge. Those are three different things, and treating them as one is how people talk themselves both into and out of good jobs.
Why this matters now
Layoffs, restructurings, and hiring freezes have made employer risk a live question rather than an academic one, and the signals are genuinely hard to read from outside. A company can post a role while planning cuts in another division. A listing can stay up for months because it is a pipeline-builder, not an opening. A five-star review burst can mean a genuine culture improvement or an internal push to fix a rating. You cannot resolve these from the outside with certainty, and a prompt that pretends otherwise is lying to you. What you can do is grade what you have, name what would settle each open question, and go get that one thing. This prompt does that in one pass and takes about ten minutes once your notes are gathered.
Act as a due-diligence analyst working for me — the candidate, not the company. I am going to paste in the raw material I gathered about one employer. Sort it, weight it, and tell me what it does and does not prove.
Company stage: [PUBLIC COMPANY / FUNDED STARTUP / PRIVATE OR FAMILY-OWNED / NONPROFIT OR PUBLIC SECTOR]
My risk tolerance right now: [LOW, I need stability / MEDIUM / HIGH, I can absorb a bad outcome]
What I am most worried about: [for example, being laid off within a year, a difficult manager, the role not really existing]
Here is what I found. Some of it is documented, some is hearsay, and I have not sorted it:
[PASTE: dated notes, review excerpts with their dates, headlines with dates, references to filings or funding announcements, anything a contact told me, and the job listing's posting and reposting history]
Step one. Sort every item into exactly one of three buckets and label it.
CONFIRMED — supported by a primary or official source I named, or by two independent sources
CONCERNING — a pattern across multiple items that is suggestive but not proven
RUMOR — single-source, undated, anonymous, or emotionally loaded with no corroboration
Step two. Run these five checks against the sorted material. Where I have not given you enough to run a check, say so plainly instead of guessing, and tell me what to go get.
1. Layoff history — how many rounds, over what period, narrow or company-wide
2. Leadership turnover — departures at or above the level of the role I would hold
3. Financial runway — for startups, time since the last raise and the stated use of funds; for public companies, the direction of the trend rather than the absolute numbers
4. Review forensics — weight recent reviews over old ones, weight repeated themes over emotional intensity, and flag any tight cluster of similar high ratings appearing close together in time. One furious review is data about one person. The same specific complaint appearing in four reviews across three years is data about the company.
5. Listing legitimacy — how long the posting has been up, whether it has been reposted, whether a hiring manager is named, and whether the described scope matches the apparent size of the team
For each of the five, give me a rating of GREEN, AMBER, or RED, one sentence of reasoning that names which of my pasted items you used, and the single most efficient next step that would move that item from CONCERNING to CONFIRMED or clear it entirely.
Standing rules. Do not add facts I did not give you. If you know something about this company from your training data, do not use it — it may be out of date, and this audit has to rest only on evidence with a date on it. Where an item I gave you has no date, say so and discount it. A short honest audit beats a padded one.
End with one line. On this evidence, is applying here worth my effort, and what single piece of new evidence would most change your answer?
How the AI reads this prompt
Practical examples from different industries
The scenarios below are illustrative. They describe how the prompt behaves, not real companies or real audits.
A startup with a strong offer and a quiet funding history.
An engineer has an offer from a company that raised a seed round three years ago and has announced nothing since. He pastes in the funding announcement date, four reviews from the past year, a note that two senior people updated their profiles to a new employer last quarter, and a comment from a former colleague that hiring had slowed. The audit marks the funding gap CONFIRMED and rates runway RED, marks the departures CONCERNING because two is a pattern but a small one, and marks the colleague's comment RUMOR. Its recommended next step is specific — ask directly in the final conversation how long the current plan is funded for, and treat a vague answer as an answer. He asks. The answer is specific and credible, and he takes the job with a clearer picture than the offer letter gave him.
A large employer with a layoff pattern and a good brand.
A marketing manager is looking at a company everyone she knows admires, but she has a nagging feeling about it. She pastes in three dated layoff headlines spanning two years, a set of reviews complaining about workload that all cluster in the six months after each round, and the fact that her prospective team has been reorganized twice. The audit does what she could not do herself, which is line up the review dates against the layoff dates and point out that the complaints are not random — they follow the cuts. It rates layoff history RED and review forensics AMBER, and notes that none of this makes the job bad, only that she should assume more work per person and negotiate accordingly. That reframing is worth more than a pass or a decline.
A posting that might not be a real opening.
A recent graduate has applied to the same listing twice in four months and heard nothing either time. He pastes in the original posting date, both reposting dates, the fact that no hiring manager is named anywhere, and that no one on the team's public profiles has the title in question. The audit rates listing legitimacy RED, is careful to say that a stale posting has several innocent explanations including a frozen requisition, and recommends the one efficient move available — find someone who currently works there and ask whether the team is actually hiring. He does. They are not. He has just recovered several hours he would have spent on a third application.
Creative use case ideas
Auditing a client or a contract opportunity.
Freelancers can run the same five checks on a prospective client with the categories swapped — payment history reports rather than layoff history, principal turnover rather than leadership turnover, and whether the project brief has been reposted across freelance platforms.
Vetting an investment or a large purchase.
The confirmed / concerning / rumor sort is not about employment at all. It works on any pile of mixed-quality evidence you have gathered under time pressure and emotional investment, which describes most consequential decisions.
Auditing your own team.
Managers can run it on internal signals — attrition dates, exit interview themes, engagement survey trends — with the same discipline about what is proven and what is felt. The output is uncomfortable and useful in that order.
Sorting evidence in a family or community decision.
Non-business use. Choosing a school, a care facility, or a contractor produces exactly the same mess of reviews, hearsay, and one strongly worded warning from a neighbor. The three-bucket sort is the whole method, and it works with no modification.
Post-mortem on a job that went wrong.
Feed in what you knew before you accepted a role that did not work out and see which bucket the warning signs were in at the time. This is how you calibrate for the next search, and it is a better use of the experience than replaying it.
Adaptability tips
Change the five checks to match your industry. Academic and research roles need grant funding cycles and principal investigator stability. Healthcare needs regulatory action and staffing ratios. Trades and construction need project pipeline and payment reputation. The three-bucket sort stays; the checks are yours to choose.
Adjust the tolerance line and watch the ratings move. Running the same evidence at low and high risk tolerance is genuinely instructive, because it shows you which amber ratings are about the company and which are about you.
Scale down for speed. Cut to three checks and skip the ratings when you are triaging eight companies and need to know only which two deserve a full pass.
Chain it to Variation 1. The snapshot brief tells you what to gather; the audit tells you what the gathering was worth. Running them in sequence on the same company takes an afternoon and produces something most other applicants will not have.
Add a devil's advocate turn. After the audit, ask the model to make the strongest honest case that every red flag has an innocent explanation. Sometimes the case is strong, and it is better to hear it before you decline than after.
Pro tips
Paste your evidence with dates attached to every single item, even approximate ones. The audit's quality is almost entirely a function of whether your input carries dates, and an undated pile produces an audit that has to hedge everything.
Ask for the audit twice — once as written, then again with the instruction to argue the opposite conclusion using only the same evidence. If the reversed version is weak, your conclusion is solid. If it is strong, you were about to make a decision on a coin flip.
Keep the RUMOR bucket rather than deleting it. Rumors that recur across three separate companies in the same industry are not rumors about companies, they are information about the industry.
Prerequisites
You need to have done the research first — this prompt grades evidence, it does not gather it. Run Variation 1 or your own equivalent before you get here. Bring dates for everything you can, the job listing's posting history, and an honest sense of your own risk tolerance, which is the input people most often fudge.
Required tools
Any general-purpose AI chat tool. A model with a longer context window helps if you are pasting a lot of raw notes, but the prompt works on a free tier if you trim your input to the items that carry dates.
Frequently asked questions
Is a long-running job posting actually a red flag?
Sometimes, and the honest answer is that you cannot tell from outside. A posting can stay up because the requisition is frozen, because the company is building a candidate pipeline, because the platform auto-renews it, or because they genuinely cannot fill a hard role. Those are very different situations and only one of them is a reason to walk away. Treat a stale posting as a question to resolve rather than a verdict, which is why the audit asks you to verify rather than concludes. The efficient resolution is almost always a person, not a document.
Can I really tell suspicious reviews from real ones?
Not with certainty, and be wary of anyone who says otherwise. What you can spot is a pattern — a cluster of short, similar, highly positive reviews appearing within a narrow window after a period of negative ones. That pattern has innocent explanations too, including a company that genuinely fixed something and asked happy employees to say so. It is worth noting that the Federal Trade Commission's rule on consumer reviews and testimonials, which took effect in October 2024, treats fabricated and undisclosed insider reviews as deceptive practices, so the incentive to fake them carries real legal risk now. Weight the pattern, do not convict on it.
Where would I even find layoff information?
In the United States, the federal WARN Act requires employers with 100 or more employees to give 60 calendar days advance written notice of plant closings and mass layoffs meeting certain thresholds, and those notices are filed with state workforce agencies, many of which publish them. Several states have their own versions with lower thresholds or longer notice periods. That covers larger layoffs at larger employers and misses smaller ones entirely, so pair it with dated news coverage and with what people's own public profiles show about when they left.
The audit came back mostly amber. What am I supposed to do with that?
Amber is the normal result and it is not a failure of the audit. Most employers are genuinely mixed, and a wall of green usually means you did not gather enough evidence rather than that the company is flawless. The useful output in an amber-heavy audit is the list of next steps — pick the two cheapest ones, resolve them, and rerun. If everything is still amber after that, the company is a normal company and your decision is about fit and compensation rather than risk.
Should I share this audit with anyone?
Keep it to yourself. It contains your unverified suspicions about a named employer alongside things people told you in confidence, and it exists to inform your judgment rather than to be presented to anyone. The output that faces the world is a better question in an interview, not a document.
Recommended follow-up prompts
The dossier matrix
— Variation 3 of this post. Once you have audited three or four companies individually, the matrix scores and ranks them against each other.
The verification-question builder
— turn every AMBER rating into a question you can ask a current employee or a recruiter without sounding like you are conducting an investigation.
The offer-stage risk recheck
— rerun the audit at the offer stage with what you learned during the interview process, which is usually more than you learned from public sources.
Tags and categories
Tags:
red flags, due diligence, employer risk, review analysis, layoff history, ghost listings, evidence grading, intermediate prompts Categories: Career and Job Search, Risk Assessment
Citations
U.S. Department of Labor, "Plant Closings and Layoffs" — Worker Adjustment and Retraining Notification (WARN) Act, 29 U.S.C. § 2101 et seq. Sets the 60-day advance notice requirement for covered employers with 100 or more employees and defines the plant closing and mass layoff thresholds that trigger it.
Federal Trade Commission, Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, effective October 21, 2024. Prohibits fake and AI-generated reviews, undisclosed insider reviews, and company-controlled review sites presented as independent.
U.S. Securities and Exchange Commission, Office of Investor Education and Advocacy, "How to Read an 8-K" — Investor.gov investor bulletin. Describes the current-report form on which US public companies disclose material events, including the departure of directors and principal officers.
The Dossier Matrix
A weighted scoring system you run on every company.
Researching one company well is a skill. Researching fifteen consistently is a system, and the difference between them is where most job searches quietly fail. By the second week the enthusiasm that carried the first three deep dives is gone, the fourth company gets a skim, the tenth gets a glance at the salary and a shrug, and the ranking in your head reorders itself according to which company answered your email most recently. That is not a shortlist, it is a recency effect with a spreadsheet next to it. What follows is the week's real deliverable — a rubric built from your own must-haves, a research routine short enough that you will actually run it fifteen times, and a scored matrix that produces two lists. The companies worth your effort, and the ones you are done thinking about.
Why this matters now
The economics of applying have inverted. When each application cost an hour of tailoring, the constraint was output. Now that it costs minutes, the constraint is discrimination — knowing which fifteen of the hundred plausible companies deserve real effort, and being able to defend the ranking to yourself at week six when you are tired. A system that produces the ranking also produces the thing you need more than the ranking, which is a documented reason for every rejection. Screening out a company with a written reason attached means you never reconsider it at 11pm, and it means that when you do reconsider it, you are responding to new evidence rather than to fatigue. This is also the artifact the rest of the series consumes — applications, interview prep, and negotiation all get sharper when they run off a scored dossier rather than a list of names.
You are going to help me build and then operate a target-company dossier system. Work in three phases and stop at the end of each phase for my input. Do not skip ahead to a later phase.
MY SPEC
Must-haves: [3-5 non-negotiables from my target-role spec, or what I know I need if I do not have one written down]
Deal-breakers: [3-5 things that would make me decline an offer outright]
Nice-to-haves: [2-4]
Constraints: [location, compensation floor, timeline, work authorization, anything else that is fixed]
Companies on my list so far: [NAMES, or write 'structure it and I will fill in names']
PHASE 1 — BUILD THE RUBRIC
Turn my must-haves and deal-breakers into a scoring rubric across these eight areas.
Financial health. Employee review patterns. Listing legitimacy. Pay transparency. Leadership signals. Growth trajectory. Interview-process reputation. Hiring pathway, meaning whether this company tends to fill roles through referral and internal networks before anything is posted.
For each area, do three things. Define what a score of 0, 3, and 5 look like in concrete, observable terms — something I could check rather than something I would have to feel. Name the source types that would let me score it. Propose a weight out of 100 that reflects my specific must-haves and deal-breakers rather than generic career advice.
Make every deal-breaker a gate rather than a weight. A company that trips a gate is screened out regardless of its composite score, because a high total should never be able to buy its way past something I said was non-negotiable.
Then show me the rubric and the weights, tell me which two areas your weighting has made decisive and why, name the area you think I will find hardest to score honestly, and ask me what to change. Stop there.
PHASE 2 — THE INTAKE PROTOCOL
Once I have approved the rubric, give me the per-company research routine. Order it so the cheapest potentially disqualifying checks come first, letting me stop early on companies that fail rather than researching them thoroughly and then rejecting them. For each of the eight areas, give me the exact question I am answering and the source type that answers it.
Cap the routine at 45 minutes per company and tell me explicitly what you cut to make it fit and what I lose by cutting it.
Then give me a blank intake block I can copy for each company — just the fields you need from me to score one company. Tell me what to write when I looked and could not find something, so that a gap is recorded as a gap rather than silently scored as a zero. Stop there.
PHASE 3 — SCORE AND RANK
I will paste back completed intake blocks, one or several at a time. For each company, score all eight areas against the rubric, showing the number and one line of reasoning that names which piece of my evidence produced it. Flag any gate tripped. Give the weighted composite.
Then rank every company scored so far into three tiers. Apply now. Apply if the top tier stalls. Screened out. For each screened-out company, give the specific reason and name the one piece of evidence that would reverse the decision.
STANDING RULES FOR ALL PHASES
Score only what I paste. Never fill a gap from your own knowledge of these companies — your information may be out of date, and a confident wrong score is worse than a blank one.
Where evidence I give you carries no date, say so and discount it.
Keep a running note of which areas I consistently leave blank, and tell me about it, because that pattern tells me where my research routine is failing rather than where the companies are failing.
After every third company I add, tell me whether the rubric is still discriminating between companies or whether the scores are clustering. If they are clustering, propose a sharper criterion for the area with the least spread.
Anything that will eventually be read by a person at these companies is mine to write. Your job here is the framework and the scoring.
How the AI reads this prompt
Practical examples from different industries
The scenarios below are illustrative composites, written to show how the system behaves. They are not real searches or real companies.
A senior engineer with twelve companies and no time.
She lists three must-haves including a hard remote requirement and two deal-breakers including on-call rotation without compensating time. Phase 1 produces a rubric that weights hiring pathway higher than she expected, and the model's explanation is that with a strict remote requirement her realistic pool is small enough that referral-first companies represent most of her real access. She would not have reached that on her own. Phase 2 puts the remote-policy check first because it is a two-minute lookup that eliminates a third of her list. By the time she reaches Phase 3 she is scoring eight companies rather than twelve, and the four she cut cost her eight minutes total.
A career changer with a shortlist scribbled on a napkin.
He has no Week 2 spec sheet, five company names, and a strong sense of what he does not want. He writes his deal-breakers first because they are clearer to him than his must-haves, which the prompt allows. Phase 1 comes back with a rubric that is honest about its own weakness — it tells him the weighting is thin because his must-haves are vague, and names the area it thinks he will struggle to score. That prompt to sharpen his own spec turns out to be more valuable than the ranking, and he goes back to Week 2's exercise before continuing. The matrix bites harder with a real spec behind it, and this is what that looks like in practice.
A hiring manager running a search while employing people.
She uses the system in the opposite direction, scoring the three teams she might move to inside her own industry, all of which she knows socially. Her problem is not lack of information but excess of unstructured impression, and the intake block forces her to separate what she has actually observed from what she has absorbed from friends. Two of the three companies score close together, the clustering warning fires after the third, and the model proposes a sharper criterion on interview-process reputation. The sharper criterion separates them immediately, and it is a distinction she had been feeling for weeks without being able to name.
Creative use case ideas
Vendor or contractor selection for a small business.
The gate-versus-weight structure, the early-exit ordering, and the null-value handling all transfer intact. Swap the eight areas for the ones that matter and the system runs unchanged.
Choosing where to apply to graduate school or a training program.
Cost, completion rate, faculty stability, where graduates actually land, and how the program is funded map cleanly onto the same architecture.
Running the matrix on your current employer as a control.
Score the place you already work using the same rubric. If it lands in your top tier, your search is about a role rather than an escape, which is a genuinely different search. If it screens out, you have written down why, which is worth having on a bad week.
Choosing a place to live.
Non-business use. Neighborhoods, school districts, and rentals produce the same problem — many options, mixed evidence, strong recency bias, and a decision that gets made emotionally at the end of a long day. The rubric-plus-gates structure is the same, and running it with a partner has the useful side effect of surfacing that you are weighting things differently.
Selecting a volunteer commitment or a board seat.
People with limited discretionary time routinely give it away to whichever organization asked most recently. A five-line rubric and one gate fixes that in twenty minutes.
Adaptability tips
Drop areas that do not apply. Someone targeting public-sector or academic roles can cut listing legitimacy and growth trajectory and add funding cycle stability and institutional accreditation. Eight areas is a starting shape, not a requirement.
Run it with fewer companies than you think you need. Four companies is enough for the rubric to prove itself, and starting small means you discover a bad weighting before you have scored twelve companies against it.
Split Phase 3 across sessions. Ask for a compact carry-forward summary of the rubric and the scores so far at the end of each session, and open the next one by pasting it back. This is what makes the system survive a search that runs for months.
Invert it as preparation. Ask the model what a company would see if it ran the same eight-area audit on you as a candidate. The answer is uncomfortable and it is excellent interview preparation.
Add a per-company confidence score alongside the composite. A company scoring 78 on rich evidence and one scoring 78 on three data points should not sit in the same tier, and tracking evidence quality separately from score is a small change with a large effect on the ranking's honesty.
Pro tips
Have the model output the rubric once, in full, and save it outside the chat. Rubric drift across long sessions is real, and pasting the canonical version back at the start of each Phase 3 session eliminates it.
Ask for the two closest pairs in your ranking and what evidence would separate them. Most of the decision value sits in the pairs that scored within a few points, and everything above and below them is already settled.
When a company you like scores badly, resist adjusting the weights. Instead ask the model which specific weight change would move it into the top tier, then decide whether you actually believe that change. Making the fudge explicit is the only defense against making it unconsciously.
Revisit your screened-out list once, at week six. Some of those rejections were made on thin evidence early in the search, and the one-line reversal condition you recorded makes the recheck take minutes.
Prerequisites
This works best with the Week 2 target-role spec sheet in hand — the must-haves and deal-breakers are its direct inputs, and the matrix is sharper when it scores against a spec you wrote deliberately rather than one you improvised at the prompt. A napkin list works to start. You will also want completed research on at least three companies, ideally run through Variation 1 and audited with Variation 2, plus a place to store the rubric outside the chat window and a willingness to run the intake routine more than twice.
Required tools
A general-purpose AI chat tool with a reasonably long context window, since Phase 3 accumulates. A tool with persistent projects or saved conversations makes the multi-session version far easier but is not required — pasting a carry-forward summary achieves the same thing. Keep the rubric and the matrix in a plain document outside the chat.
Frequently asked questions
Is eight areas too many to research for every company?
It would be if you researched all eight equally, which is why Phase 2 caps the routine and orders it for early exit. In practice most companies are resolved in the first two or three checks, and only the ones that survive get the full pass. If you find yourself running all eight on every company, your ordering is wrong — move whatever most often eliminates a company to the front and tell the model to rebuild the sequence around it.
How do I score pay transparency when the posting has no salary range?
The absence is the score, and it is worth knowing that in some jurisdictions it is also a compliance question. Several US states and cities require employers to disclose a compensation range in job postings, including Colorado under its Equal Pay for Equal Work Act, New York City under Local Law 32 of 2022, and California under SB 1162. A posting for a role in a covered location with no range is a small, concrete data point about how the company handles rules that are inconvenient for it. Outside covered jurisdictions it means much less, so score it against local norms rather than absolutely.
What if my rubric ranks a company first and my gut says no?
Take the gut seriously and interrogate it rather than overriding either one. Usually it means one of two things — a criterion you care about is missing from the rubric, or you have evidence you never wrote into an intake block because it felt too subjective to record. Both are fixable. Add the missing area, or add the evidence, and rerun. A ranking that disagrees with you is doing its job by making you articulate why.
Can I run this on a free tier?
Phases 1 and 2 comfortably. Phase 3 accumulates context as companies pile up, and long free-tier sessions will start losing the early rubric. The fix is the carry-forward summary — end each session by asking for a compact restatement of the rubric, the weights, the gates, and the scores so far, and open the next session by pasting it back. That keeps the system running indefinitely regardless of context limits.
Won't a system this structured make me miss a good opportunity that does not fit the criteria?
It can, and it is worth being honest about that trade. The system optimizes for not wasting effort, and any filter that saves you time will occasionally filter something you would have liked. Two mitigations help. Keep the screened-out list with reversal conditions attached, so a good opportunity that arrives from an unexpected direction can be reconsidered against evidence rather than mood. And keep one slot in your week for something entirely outside the matrix — the referral from a former colleague that fits nothing on your list is a real category, and no rubric will ever surface it.
Recommended follow-up prompts
The application prioritizer
— Week 5 of this series consumes this matrix directly. Feed the top tier into the application workflow and let the ranking decide the order.
The interview prep loader
— Week 6 draws on the dossier. Take one company's completed intake block and turn it into the specific questions and the specific unknowns you should be working on before the interview.
The negotiation context brief
— Week 7 leans on what you know about the company's position. The financial health and growth trajectory areas of the dossier are exactly the inputs that conversation needs.
Tags and categories
Tags:
scoring rubric, decision matrix, target company list, weighted criteria, multi-step prompting, research systems, job search strategy, advanced prompts Categories: Career and Job Search, Decision Systems
Citations
Colorado Department of Labor and Employment, Equal Pay for Equal Work Act — job posting compensation disclosure requirements. Requires covered employers to include compensation and benefits information in job postings.
New York City Commission on Human Rights, Local Law 32 of 2022 — salary transparency in job advertisements. Requires covered employers to state a good-faith salary range in advertised positions.
California Labor Code § 432.3, as amended by SB 1162 (2022) — pay scale disclosure in job postings for covered employers.
Anthropic, prompt engineering documentation — docs.claude.com. Guidance on multi-step prompting, role assignment, structured output, and defining evaluation criteria within a prompt.
Which of the three should you use?
The three prompts are not one method at three lengths, and the clearest way to see the difference is to ask what each one takes as input. The Snapshot Brief takes a company name and produces questions. The Red-Flag Audit takes evidence and produces judgments. The Dossier Matrix takes criteria and produces a ranking. They operate on different material at different stages, and running them out of order mostly does not work — the audit has nothing to grade if you have not researched, and the matrix has nothing to score if you have not audited.
Choose by where you are stuck. If you do not know what to look for, start with Variation 1; it is the only one that will teach you the shape of employer research from nothing, and a reader who has never done this before should not skip to the matrix, which will produce a beautifully structured system with no evidence in it. If you have researched a company and cannot tell whether what you found is bad, Variation 2 is the whole answer, and it is also the one worth running on its own — plenty of readers will use the audit and nothing else, and they will be better off for it. If you have five or more companies and a real sense of what you need, go to Variation 3, and expect the rubric-building phase to be more valuable than the ranking that comes out of it.
They overlap in one deliberate place, which is the discipline about evidence. All three refuse to let the AI tell you facts about a named company, for the same reason — a language model does not know whether that company laid people off last month, whether the posting you are looking at is real, or who runs the department now, and a confident wrong answer on any of those costs you either a good job or a bad one. What the AI is genuinely good at is the part underneath the facts. It builds the questions, defines the standards of evidence, holds the rubric steady across fifteen companies when your attention will not, and refuses to let a strong salary quietly outweigh something you called non-negotiable. That is what an analyst does. The investigating stays yours, which is also why the findings will hold up when someone asks you about them across a table.
TAGS:
