Guide

Research a Public Company as Four Separate AI Jobs

Tool-agnostic
Intermediate
8
min
Account Executive
Sales Development
ChatGPT
Claude
Perplexity
October 25, 2026

Most AI account research starts with one prompt: research this company. For a strategic public company, that single request asks one run to do four different jobs, each with its own primary source, recency window, and failure mode. The answer reads smoothly and blends them together: an old press release sits beside last quarter’s commentary, a boilerplate risk factor reads like live pain, and an executive’s priorities are guessed rather than quoted.

This guide separates the work into four jobs—account plan, latest earnings, annual and quarterly filings, and executive material—runs each against its own sources, and merges only the claims that survive verification. The approach comes from a practice sales practitioners have shared publicly for running account research in Perplexity as separate prompts; we could not relocate a stable link to the original post, so we credit it generally. Nothing here depends on one tool: any approved assistant that can search the web and cite its sources will do.

When to use this guide

Use it for strategic, publicly listed accounts before a first executive meeting, an account plan, or a re-engagement. The filings job assumes a company that files with the U.S. Securities and Exchange Commission. For companies listed elsewhere, substitute that market’s regulatory filings. For private companies, drop the filings and earnings jobs rather than letting a model estimate figures that are not public.

For long-tail accounts, a single sweep is still the right tool. The 15-Minute Account Research Brief uses one comprehensive prompt to decide whether an account deserves more effort. This guide is what that extra effort looks like.

Why separate the jobs

Each job has a different best source and a different way of going wrong:

JobPrimary sourceTreat as currentTypical failure
Account planCompany website, investor-relations pages, official announcementsLast 6–12 monthsA company profile with no seller decision behind it
Latest earningsThe earnings release and the call transcriptMost recent quarterPrepared remarks quoted as priorities; quarters mixed together
10-K and 10-QThe filing itself, retrieved from EDGARLatest annual and quarterly filingRisk-factor language read as confirmed pain
Executive materialThe executive’s own public statements, interviews, and official bioDated statements, newest firstPersonality inference, or research on the wrong person

The recency windows are a Promptifi working rule, not a regulatory standard. Adjust them to your sales cycle, but state them in every job.

Step 1: Write one shared header

All four jobs start from the same header so their outputs can be merged later. Write it once and paste it above each job:

Illustrative header — paste above each job

“Company: [COMPANY NAME AND TICKER]. Fiscal year end: [MONTH]. I sell [WHAT YOU SELL] to [BUYER ROLES]. This research supports one decision: [SELLER DECISION]. Treat as current: [RECENCY WINDOW FOR THIS JOB]. Cite every factual claim with its URL and publication date. Label each line as sourced fact, inference, or unknown. If the sources do not establish something, write NOT FOUND instead of estimating it. Cover only this research job; do not summarize the others.”

The last sentence matters. Without it, each run drifts back toward a full company profile and the separation is lost.

Step 2: Job one, the account plan

Ask for the business model, the changes inside the stated window, and the seller-relevant hypotheses those changes suggest, with a source table. Build a Comprehensive Account Brief with Deep Research is built for this job and includes an optional research-plan step for important accounts. Keep its output as job one’s ledger, and do not let it summarize earnings or filings it has not read.

Step 3: Job two, the latest earnings

Work from the primary material, not news coverage of it. Investor.gov notes that many companies announce quarterly and annual results “simultaneously in a press release and an 8-K,” and that the 8-K often announces an upcoming earnings call (Investor.gov: How to Read an 8-K). Get the release and, where the company publishes one, the call transcript, and give them to the model directly.

Extract Buying Signals from an Earnings Call Transcript handles this job. It works the analyst Q&A separately from the prepared remarks and flags questions management did not answer directly.

Step 4: Job three, the 10-K and 10-Q

Retrieve the filing yourself. The SEC’s EDGAR company search finds a company’s filings by name or CIK number, and full-text search covers filings from the last four years (SEC: How Do I Use EDGAR?). Investor.gov describes the annual 10-K as offering “a detailed picture of what the company does, and the risks it faces,” and points readers to the Business section, Risk Factors (“generally listed in order of importance”), and Management’s Discussion and Analysis (Investor.gov: How to Read a 10-K). The quarterly report, Form 10-Q, carries the full financial statements that follow an earnings 8-K, so check for one newer than the 10-K.

Extract Sales Intelligence from a 10-K Filing works only from the filing material you paste and keeps what the filing says apart from what you read into it. Treat a risk factor as a disclosure, not as confirmed pain: it tells you what the company chose to disclose as a risk, not that the company is acting on it. Look for evidence from the other jobs before you build on it.

Step 5: Job four, executive material

Research the named executives last, once you know what the company is prioritizing. Confirm each person’s current title on the company’s own site first, so the brief is about the right person. Limit the job to their own public, professional statements: interviews, conference remarks, published articles, earnings-call answers, and the official bio. Build a Buyer Intelligence Brief on Any Executive separates verified facts from assumptions and flags what it could not confirm. Leave personality, politics, and private life out entirely.

Step 6: Merge only what survives verification

Now combine the four ledgers into one. For each claim, keep the job it came from, the source and date, and its label. Then:

  1. Open the sources behind every claim you plan to repeat to a buyer.
  2. Where two jobs disagree—say, an announcement in job one and a more cautious answer on the earnings call in job two—keep both and note which is newer.
  3. Move anything marked NOT FOUND to an unknowns list, and turn the important ones into discovery questions.
  4. Write one or two hypotheses that connect evidence from at least two jobs, labeled as inference.

The labeling rules are in Facts, Inferences, Unknowns, and Recommendations.

Optional: run one job twice

For a claim that will carry a meeting, run the same job in a second assistant or a deeper research mode and compare the claims, not the prose. Agreement on sources raises confidence; disagreement tells you what to verify. Why AI research changes with reasoning mode shows the comparison table to use.

Optional: put the filings job on a schedule

For a short list of named public accounts, the 10-K and Earnings Watcher automation checks for new filings and earnings content on a cadence and returns quotable passages with citations. Spot-check what it returns before you repeat it to a buyer.

Worked example

Fictional account: Halvorsen Freight, a U.S.-listed logistics company. A single sweep produced a confident line: “Halvorsen is expanding into Europe and investing heavily in automation.” Run separately, the four jobs told a more useful story:

  • Account plan: the European expansion came from an announcement older than the recency window, so it moved to background.
  • Earnings: in the analyst Q&A, the CFO said automation spending would be phased over several years. That answer is newer than any other evidence on the topic.
  • Filings: the latest 10-K lists labor availability and warehouse throughput among its risk factors. Disclosed risk, not confirmed pain.
  • Executive: the COO’s recent conference remarks focused on reducing manual exception handling in existing warehouses.

The merged hypothesis, labeled as inference: “Halvorsen may be prioritizing throughput in its existing sites over new automation spend this year.” The question that tests it: “How are you balancing throughput in today’s warehouses against the phased automation plan you described on the last call?” The sweep’s “investing heavily” claim never reaches the email.

Common failure modes

  • One prompt for a strategic account: fine for triage, not for an executive meeting. Separate the jobs.
  • Coverage instead of source: the earnings job summarizes articles about the call. Give it the release and transcript.
  • Disclosure read as pain: a risk factor is a disclosure. Look for other evidence that the company is acting on it.
  • Stale filings: an annual report late in the fiscal year is old news. Check for a newer 10-Q.
  • Mind-reading: the executive job guesses motives. Keep it to dated public statements and turn the rest into questions.
  • Merging before verifying: four unverified ledgers combine into one confident, unverified brief. Verify, then merge.
  • Moving account context between tools: do not copy sensitive account information into unapproved tools just to run a job somewhere else.

Put it to work

Run the four jobs with Build a Comprehensive Account Brief with Deep Research, Extract Buying Signals from an Earnings Call Transcript, Extract Sales Intelligence from a 10-K Filing, and Build a Buyer Intelligence Brief on Any Executive, then carry the merged evidence into the source-grounded executive POV guide.

Prerequisites

  • A publicly listed target account and its ticker
  • The seller decision the research supports
  • An AI assistant with web search and citations that your organization permits
  • Access to the company’s investor-relations pages and SEC EDGAR
  • The latest earnings release and, where published, the call transcript
  • The names of the executives you plan to engage

Troubleshooting

If a job’s output drifts into the other jobs, repeat the “cover only this research job” line and give it only that job’s sources. If the earnings job cannot find a transcript, give it the release and mark call-specific findings NOT FOUND rather than letting it reconstruct the call from coverage. If the filings job quotes language you cannot find in the filing, discard the claim and rerun the job on pasted sections only. If two jobs keep contradicting each other, keep both claims and make the contradiction your first discovery question.

Sources

Investor.gov (SEC) — How to Read a 10-K

Investor.gov (SEC) — How to Read an 8-K

SEC — How Do I Use EDGAR?

OpenAI Help Center — Searching the web with ChatGPT

Anthropic Help Center — Enable and use web search

Perplexity Help Center — What is Research mode?

Method origin: a practice shared publicly by sales practitioners for running public-company account research in Perplexity as separate prompts. The original post could not be relocated, so it is credited generally. All pages above verified 26 October 2026.