How to Research a Prospect Using AI Before a Sales Call

Why Generic Prospect Research Fails

Generic research fails because it answers a question the buyer never asked. They know what their company does. What they want to find out in the first four minutes is whether you understand what their week actually feels like.

Ask most buyers what kills a discovery call fastest and you get the same answer: being told about their own company. The moment a rep opens with a summarised About page, the meeting becomes a demo request in the buyer’s head. So the failure is not that the AI produced bad writing. It produced accurate, useless writing.

The second version works because it removes the AI’s freedom to be vague. It cannot return “they care about growth and efficiency” when the format demands a date, a source tag, and a link between two specific things. Constraint is what turns a summariser into a researcher — the same principle behind telling the model what it must not do.

The Four-Pane Pre-Call Brief

The Four-Pane Pre-Call Brief is a fixed four-part output you ask an AI tool to fill before any first meeting: Ground, Signal, Pressure, Angle. It is a shape, not a script — the same four panes work for a fifty-person agency and a listed manufacturer, and they work in whichever tool you already pay for.

Pane One: Ground

Ground is the boring pane, and skipping it is why briefs collapse. It holds business model, who they sell to, rough size, and how they make money — each item tagged as sourced or unverified so you know which sentences you are allowed to say out loud.

Pane Two: Signal

Signal is anything dated from the last six months: a funding round, a new market, a leadership change, a product launch, a merger. Undated signal is not signal. If your tool cannot attach a date and a source to an item, that item belongs in the bin, not in your opening.

Pane Three: Pressure

Pressure is where most reps stop short. A VP of Customer Success and a VP of Finance at the same company are under completely different measurement. Naming the metric that lands on this person’s desk is what separates a tailored call from a personalised one, and it depends entirely on how much situational context you give the model about the role and industry.

Pane Four: Angle

Angle is the output. Three questions, each connecting one Signal item to one Pressure item, none of them answerable from the company website. These are raw material for the conversation rather than a script — turning them into a full call plan is its own discipline, covered in our guide to preparing discovery call questions using AI.

How to Research a Prospect Using AI Before a Sales Call

Three steps, roughly ten minutes once you have done it twice. Gather, run, verify. The temptation is to skip the third one because the output looks finished — resist it, because polished wrong is more dangerous on a call than obviously thin.

Do not hand over a list of links and hope. Select the text on the page, copy it, paste it in. This sounds primitive and it is the single most reliable habit in the whole workflow — pasted text cannot be blocked, rate-limited, or half-loaded, and you can see exactly what the model is working from.

• The company’s About and product pages — the text, not the URL.
• Their newsroom or press page, plus a quick news search on the company name.
• Two or three live job postings from their careers page.
• Anything public the prospect has written or said themselves.

LinkedIn is the usual sticking point. Profile pages sit behind a login wall, so pointing a chatbot at a profile URL tends to produce either a refusal or, worse, a confident summary of a page it never read. Copy the visible About and Experience text and paste that instead.

One prompt, one output. The role and format instructions do the heavy lifting here — this is the Logic System applied to research rather than writing, with the four panes acting as the Format element.

You are a B2B research analyst preparing me for a discovery call. I sell [one sentence: what you sell and to whom].

The call: [prospect name], [job title] at [company name]. Meeting length: [30] minutes.

Below is the raw material I collected. Work from this material and live web search only. Do not fill gaps with general background knowledge about the company.

--- COMPANY PAGE TEXT ---
[paste]

--- RECENT NEWS AND ANNOUNCEMENTS ---
[paste]

--- JOB POSTINGS ---
[paste]

Return exactly four sections, in this order:

GROUND — five facts about their business model, customers and rough size. One line each. Tag each line SOURCED or UNVERIFIED.

SIGNAL — three things that changed in the last six months, each with a date and a source link. If you find fewer than three dated items, write "No verified signal found" instead of filling the gap.

PRESSURE — three metrics or operational problems a [job title] in this industry is measured on right now. Label these as hypotheses, not facts.

ANGLE — three open questions, each connecting one SIGNAL item to one PRESSURE item. No question that could be answered from their homepage.

Rules: never invent a funding round, acquisition, executive name or customer. Keep the whole brief under 400 words.

Open every source link behind every dated claim. Not a sample of them — all of them. This takes about five minutes and it is the entire reason the brief is safe to use, because a fabricated acquisition delivered confidently to the person who works there is not a small error. It is the end of the meeting.

Triage helps: Ground facts tagged UNVERIFIED never get spoken aloud, Signal items get checked against their source, and Pressure is a hypothesis you test with a question rather than assert. Our verification system for AI research covers the full pass in detail.

Review the brief you just produced. For every claim in the GROUND and SIGNAL sections, return a table with three columns:

CLAIM | SOURCE URL | CONFIDENCE

Confidence is one of: CONFIRMED (you can point to a live source that states this), INFERRED (you reasoned it from something else), or UNSUPPORTED (you cannot locate a source).

Then list separately every claim rated UNSUPPORTED, and remove it from the brief.

Do not defend a claim you cannot source. Removing it is the correct answer.

Three Deeper Passes for Bigger Deals

The four panes cover most first calls. When the deal is large enough to justify another fifteen minutes, three extra passes add depth that competitors almost never bring.

The Filings Pass

If the company is publicly listed, its annual report is a stated list of what leadership is worried about, written by them, under legal obligation to be accurate. You can pull filings free from the SEC’s EDGAR full-text search, then hand the document to a tool built for long files — the workflow in our guide to analysing long documents with Claude applies directly here.

I have uploaded [company name]'s most recent annual report. Answer only from this document.

1. STRATEGIC PRIORITIES: the three initiatives leadership states most clearly. Quote the phrasing they use.
2. STATED RISKS: three risks or headwinds named in the risk factors section.
3. INVESTMENT DIRECTION: where spending is described as increasing or decreasing.

For each item, give the section or page it came from. If the document does not address one of these, say so rather than inferring from elsewhere.

The Job Postings Pass

Job adverts are the least guarded document a company publishes. They name the tools in use, the team that is growing, and — read carefully — the problem the new hire is being brought in to fix. Three postings from one department will tell you more about operational friction than any press release.

Below is the text of three current job postings at [company name], all in [department].

[paste]

Extract:
1. TECH STACK: every named software, platform or tool.
2. FRICTION: the operational problems these roles are being hired to solve. Quote the phrasing that implies each one.
3. DIRECTION: what the hiring pattern suggests the team is trying to build or fix.

Only list a tool if it is named in the text. Do not infer a stack from the industry.

The Competitive Angle Pass

When you know the prospect already runs a competitor’s product, the goal is not ammunition. It is a set of neutral questions that let the buyer discover a gap themselves — the same logic behind building a sales battlecard with AI.

Our prospect [company name] currently uses [competitor product] for [function]. We sell [your product], which differs mainly by [key difference].

Search recent public user feedback about [competitor product] from the last twelve months and return:

1. Three recurring complaints, each with a source link and rough date.
2. Any pricing or feature changes they have announced.
3. Three neutral, non-leading questions I can ask that would help the buyer notice these limits themselves if they are experiencing them.

Do not write anything disparaging about the competitor. If a complaint is not sourceable, leave it out.

Choosing the Right AI Tool

Any of the four major assistants will run this workflow. What differs is which source material each one handles most comfortably, and the honest answer is that your existing subscription is almost always the right choice.

On cost: paid tiers sit around twenty dollars a month across the board, and Anthropic’s published plan comparison is a fair reference point for what that buys. But entry-level tiers now exist well below that, and the free tiers have quietly become capable — free Claude and free Gemini both include web search, so the brief itself does not require a subscription. Volume does. If you are running four briefs a week you will meet a usage limit mid-brief eventually, and that is the real argument for paying.

For live signal work specifically, tools that show their sources inline save the most time on the verification pass, which is worth reading about in our comparison of Perplexity and ChatGPT for deep research.

Keeping Prospect Data Out of Trouble

Prospect research means pasting other people’s information into a third-party system, and most sales teams have a policy about that whether or not anyone has read it. The line worth knowing is not free versus paid.

Worth reading alongside this: our practical guide to using ChatGPT safely at work goes through the settings themselves.

When the Brief Goes Wrong

Four things go wrong often enough to be predictable. Each has a fix that lives in the prompt rather than in a different tool.

The first version is not just wrong, it is unrecoverable in the moment. Fabricated company events are the most common failure in this workflow and they come from the same root cause every time, which we cover in depth in our explanation of why AI tools invent confident details.

What AI Cannot Tell You

The brief gets you to the call informed. It does not get you through the call, and being honest about the gap is what keeps the tool useful rather than dangerous.

There is also a timing limit worth naming. Anything that happened in the last week or two may not be indexed yet, and a model that reports nothing recent is often right that nothing is findable rather than right that nothing happened. Treat silence in the Signal pane as missing data, not as an answer.

Run the four panes twice and the shape stops feeling like a process. Stripped right back, how to research a prospect using AI before a sales call amounts to a fixed shape, filled from material you gathered yourself, checked before you speak. Most reps who stick with it save the brief prompt somewhere they can reach in thirty seconds, fill three brackets, and walk in with one question nobody else on that buyer’s calendar could have asked.