Consultants use AI for client research by giving a research mode (ChatGPT deep research, Gemini Deep Research, Claude Research or Perplexity) a fixed brief: what the company sells, to whom, what has changed recently and what pressures follow. Demand a source for every claim, check the important ones yourself, then turn the findings into three hypotheses and five questions. Budget 20 to 45 minutes.
The catch is that for small private companies, AI fills gaps with plausible guesses: a revenue figure, a headcount, a founder's background borrowed from someone with the same name. Walking into a call with one of those guesses is worse than knowing nothing. The research is there to help you ask better questions, so treat every unsourced line as something to test on the call rather than something to say.
Scale the prep to what the call could be worth
Not every discovery call deserves an hour of research. Decide the depth before you open a tool, or you will either underprepare for the big ones or burn afternoons on enquiries that were never going anywhere.
| Type of call | Time budget | What you produce |
|---|---|---|
| Inbound enquiry from a small firm, fit unclear | 10 to 15 minutes | Five-line scan and two questions |
| Referral or a firm in your core sector | 25 to 45 minutes | One-page brief, three hypotheses, five questions |
| Competitive pitch or a large engagement | 60 to 90 minutes, split over two sittings | Brief plus a view on their market, competitors and likely objections |
The middle row is where most consultants' calls sit, and where the routine below is aimed.
A quick sum shows why the tiers matter. Say a month brings eight discovery calls: three inbound enquiries at 15 minutes each, four referrals at 40 minutes and one competitive pitch at 90. That's 45 plus 160 plus 90 minutes, just under five hours. Giving all eight the full treatment of an hour and a half would take twelve, most of it spent on inbound calls that were never going to become work. Decide the tier when the booking arrives, and put the time in the calendar next to the call.
The fifteen-minute scan works for any prospect
For a quick scan, a normal chat with web search switched on is enough. Paste the company name and website, and give it a narrow job:
I have a discovery call with [company], website [URL], on [date].
Using their website and other public sources, answer in no more than
ten lines, with a source link after every line:
1. What they sell and to whom (their words, not yours)
2. Rough size, only if a source states it (otherwise write "not found")
3. Anything that changed in the last 12 months: new services, new hires
at senior level, office moves, acquisitions, rebrands
4. How they describe what makes them different
5. Job adverts currently open, and what those roles suggest
Do not estimate figures. Do not describe any individual beyond their
public job title.Point 5 is the one generic research misses. Open roles say more about a firm's current pressure than any "About us" page. A small IT support firm advertising for two service desk engineers and a service delivery manager at once is probably struggling with ticket volume or response times, whatever its website says about proactive support.
An illustrative scan result for a small web design studio, and what a quick read changes:
1. Designs and builds websites for hospitality and retail brands. [site]
2. Approximately 15-20 employees. [business directory listing]
3. Launched a website care plan in March; new head of client services
joined in May. [site news page; company social post]
4. "Design that sells, built to be looked after." [site homepage]
5. Hiring a WordPress developer and an account manager. [careers page]Line 2 goes: a directory estimate isn't a source, and "15-20" may be three years old. Line 3 is the gold. A new care plan plus a new head of client services plus an account manager vacancy points to a studio trying to move from one-off builds to recurring revenue, and that is worth one question on the call. Five lines, two minutes to check, and the call has a direction.
When the fee justifies it, go deeper with a research mode
Research modes run many searches, read the pages and write a sourced report. ChatGPT's deep research comes with a monthly allowance that depends on your plan, Claude's Research needs a paid plan and web search switched on, Gemini offers Deep Research with higher limits on Google AI Pro and Ultra, and Perplexity Pro ($20 a month) is built around cited answers. If you already pay for one assistant, use its research mode before buying another; Perplexity and ChatGPT compared for business research covers when a second tool is worth it.
A deep-research brief should tell the tool what you'll use the output for. That changes what it looks for:
I'm a consultant specialising in [your area]. I'm preparing for a
discovery call with [company] ([URL]), a [sector] business. I want to
understand the operational and commercial pressures they are likely to
face, so I can ask sharp questions. Research:
- Their services, pricing signals and the clients or sectors they name
- Changes in the last 18 months (people, services, locations, ownership)
- What their clients say publicly (reviews, testimonials, case studies)
- Pressures common to firms of their type and size right now
- Two or three named competitors and how they position themselves
Separate facts (with sources and dates) from inferences (label them).
Flag anything where sources disagree. Ignore personal social media.Ask for dates on every source. Research tools surface a 2022 press release as readily as last month's news, and "they recently opened a second studio" is embarrassing when it happened three years ago and closed last spring.
From findings to hypotheses you can test on the call
A brief full of facts still leaves you starting the call with "So tell me about your business". The step that makes research useful is converting it into hypotheses: short, falsifiable guesses about what is going on. Use this shape:
Because [evidence, with source],
they may be dealing with [problem],
which would show up as [observable symptom].
Question to test it: [open question that doesn't lead]Filled in for an illustrative PR consultancy that has posted three account executive vacancies in two months:
Because they have advertised three account executive roles since July
(careers page) and two senior staff left for a rival (their posts),
they may be dealing with junior-heavy teams on client accounts,
which would show up as seniors doing reporting and admin late at night.
Question to test it: "Who pulls together client coverage reports at
the moment, and how long does it take them each month?"The question doesn't mention the departures, doesn't assume the answer, and gives the prospect room to say "we've already fixed that", which is also useful information.
Then ask the AI to help, but keep the reasoning yours:
From the facts in this brief (ignore the inferences), suggest five
hypotheses in the format below about problems this business may have
that relate to [your service]. For each, name the specific fact it rests
on. Then rank them by how easy they would be to test in a 45-minute call.Discard any hypothesis that rests on an inference rather than a fact, and any that you couldn't raise without sounding as if you'd been digging through their bins. Three good ones are plenty.
An illustrative batch of five for the video production company in the prep pack below, and what survives:
- Producer and coordinator vacancies plus delivery-date reviews suggest scheduling is stretched. Keep: rests on two checked facts.
- The animation launch suggests they need a new pricing model for mixed live-action and animation work. Keep, reworded: the fact is the launch; the pricing need is a guess, so it becomes a question about how they quote mixed projects.
- Reliance on a few large clients leaves them exposed. Discard: the brief contains no client concentration data; the model inferred it from the testimonials page.
- A competitor's fixed-price packages are putting pressure on their pricing. Keep for later: true as a fact about the competitor, but only raise it if they bring up pricing.
- Staff turnover among editors is high. Discard: it rests on one job advert that could be growth, and asking about turnover on a first call feels like prying.
When there's almost nothing about them online
Many good prospects have a five-page website, a dormant social account and no press coverage. This is exactly when research tools are most tempted to pad, so change the question rather than pushing harder for company facts.
- Research the category instead. Ask what firms of this type and size typically struggle with: a six-person translation agency's usual pinch points around project management, freelancer capacity and pricing per word. Label the output as sector context, not facts about this firm.
- Read what they chose to publish. Their services page, the clients they name and the language they use about themselves tell you how they want to be seen. Ask the AI to summarise their positioning in three sentences and quote the phrases it relies on.
- Use the enquiry itself. Whatever they wrote when they booked the call is the richest source you have. Paste it in and ask what it implies they have already tried, and what they seem to be worried about.
- Accept a thin brief. Two confirmed facts and one good question beat a page of confident filler. A short brief is also a reason to spend more of the call listening.
The category and enquiry techniques both work on the translation agency. Asked for sector context, an assistant returns, illustratively: "Small translation agencies commonly struggle with freelancer availability at short notice, with project managers acting as the bottleneck for quoting, and with clients pushing for machine-translation discounts. Typical rates are $0.10 to $0.15 per word." The first sentence is useful sector context, clearly labelled. The rate range has no source and may be wrong for their languages and clients, so it comes out; quoting a rate to an agency owner is the fastest way to show you don't know their market.
Their booking form said: "We've tried two project management tools and nobody sticks with them. We're growing and it feels chaotic. Want to talk about getting organised before we hire again." Asked what that implies, the assistant suggested they had tried a tool-first fix that didn't hold, that growth is outpacing process, that hiring is on hold until something changes, and "an estimated team of 8 to 12". The first three are fair readings of their own words and make good questions ("What made people stop using the second tool?"). The headcount is a guess dressed as an inference. Delete it, and ask.
The prompt change that matters most here is permission to stop: tell the tool that "not found" is a good answer and that you would rather have three lines than ten guessed ones. Without that instruction, most assistants treat an empty result as a failure and fill it.
A ten-minute fact-check before anything reaches the call
Consultants lose credibility in discovery calls through small wrong facts far more than through missing ones. Before the call, check these by clicking through to the source yourself:
- Right company. Common names collide. Check the website domain and the registered name match what the AI researched.
- Dates. Anything you might mention needs a date inside the last 12 months, or you phrase it as history.
- Numbers. Headcount, revenue and client counts must trace to a source that states them. If the source is a directory estimate, don't quote it.
- People. Confirm the person you're meeting still holds the role the AI gave them. Job titles in research output are often a year stale.
- Quotes. If the brief quotes the company, find the quote on the page. Paraphrases presented as quotations are common.
The mistake this catches most often is a merged company. In one illustrative case, a research report on a small IT support firm listed a managed-security launch, forty staff and an office in a second city. Only the first item was theirs; the rest came from a larger firm with a near-identical name. The tell was a source link on a different domain. Opening the call with "congratulations on the new office" would have ended the consultant's credibility in the first minute.
For a fuller method, the routine in checking sources and citations in AI research applies here unchanged.
Where the line sits on researching people
Researching a business is normal preparation. Profiling individuals is not, and prospects notice the difference. Keep to what a sensible person would expect a consultant to know: the job title, what they have said publicly in a professional capacity (talks, posts on the company's channels, interviews), and their firm's public announcements.
Stay away from personal social accounts, family details, home locations and anything that would make you say "I saw that you..." in a way that unsettles. Data-protection law in many places also limits building profiles of individuals without a good reason, so don't store personal research beyond what the engagement needs. And if the prospect has already sent you a confidential brief or tender document, don't paste it into a consumer AI plan with the model-training setting switched on.
A prep pack for a 25-person video production company
Here is how the middle-row routine plays out in an illustrative case. A solo operations consultant has a referred call with a 25-person video production company that makes corporate and brand films. Budget: 40 minutes.
- Scan (10 minutes). The quick prompt returns: they sell brand films and event coverage to mid-sized companies; they launched an animation service eight months ago; they are advertising for two producers and a post-production coordinator.
- Deep research (15 minutes, mostly waiting). The report adds client testimonials praising creative quality but two public reviews mentioning missed delivery dates, a competitor that now advertises fixed-price packages, and a line saying they "recently expanded to a second studio". Its source is a news item from four years ago.
- Fact-check (8 minutes). The second studio is dropped (old, and the website lists one address). The producer adverts are confirmed live. The reviews are real but only two, so they become a question, not a claim.
- Hypotheses (7 minutes). Hiring producers plus a coordinator, alongside reviews about delivery dates, suggests production scheduling is stretched since the animation launch. Test it with: "How has taking on animation changed the way you plan the production calendar?"
The one-page brief that goes into the call has five facts with dates, three hypotheses with their test questions, and one line on the competitor's fixed-price packages in case pricing comes up. What it doesn't have is anything the consultant would be uncomfortable saying out loud.
Using the brief in the first ten minutes of the call
Research goes wrong in the call itself when it turns into a recital. Nobody wants to hear their own company described back to them for five minutes. Open with one observation and a question, and let them correct you:
- "I noticed you're hiring two producers at the moment. Is that growth, or filling gaps?"
- "Your site talks a lot about the animation work. How much of the business is that now?"
Before and after, for the video production call in the example above. The recital version: "So I had a look, and you're a 25-person studio, you do brand films and events, you launched animation eight months ago, you've got producer roles open, and I saw a couple of reviews about deadlines..." The prospect is now defending their reviews. The better version: "You've added animation this year and you're hiring producers. What's that done to how you schedule work?" Same research, one sentence, and they start talking about the problem.
Their corrections are the most valuable output of the whole exercise, because they tell you where the public picture and the real business differ. Note them straight after the call. If your next step is a proposal, feed the corrected brief into the proposal routine for consultants rather than the original research. For prospecting before anyone has booked a call, finding B2B prospects with AI research tools uses a similar method at list scale, and competitor research in an afternoon helps when a prospect asks how they compare.
A research log that makes each call faster
Save every brief, with the corrections from the call, in a Project or folder organised by sector. After ten calls with web design studios or PR consultancies, you have a sector primer: the pressures that keep coming up, the competitors that keep being named, the hypotheses that proved right. Paste that primer into the next deep-research prompt as context and the tool starts from your hard-won knowledge instead of from zero.
A filled-in primer entry after ten illustrative calls with web design studios might read: pressures that recur are moving from one-off builds to recurring care plans (six of ten), scope creep on fixed-price builds (seven of ten) and a founder who still approves every design (five of ten); the competitors named most often are two template-based platforms and one larger agency; the hypothesis that proved right most often is "new account manager roles mean the founder is trying to step back from client contact". Counts matter here. "Six of ten" tells you how much weight a pattern deserves on the next call; "common" tells you nothing.
Keep the log about businesses and patterns, not individuals, and prune it once a year. A two-year-old brief about a prospect you never won is clutter at best and a data-protection question at worst.
Further reads
- The AI Tool Stack a One-Person Consultancy Actually Needs — Which research and chat tools a solo practice needs to pay for.
- How Much Time Can AI Save a Solo Consultant Each Week? — Where research prep sits among a consultant's weekly time savings.
- Reusing Past Client Work With AI Without Leaking Client Data — Reuse earlier sector research without leaking a past client's data.
- How to Catch Outdated Information in AI Answers — Spot stale facts before you quote them to a prospect.
- How to Run a SWOT Analysis of Your Business With AI — A structured way to pressure-test your hypotheses after the call.
- How to Use ChatGPT as a Business Adviser Without Being Misled — The limits of AI reasoning when you move from research to advice.
- AI for Recruitment Business Development: Winning New Clients — A weekly business-development routine for agencies: hiring signals, five-minute account briefs, outreach that names the vacancy, and consented spec profiles.
- How Coaches Use AI to Qualify Discovery Call Bookings — Set fit criteria, ask four questions before the calendar, let AI sort and brief each booking, and send people who aren't ready somewhere useful.
- Is Perplexity Pro Worth Paying For in a Small Business? — Who in a small business gets $20 a month of value from Perplexity Pro, how to test it against the assistant you use now, and the privacy setting to change.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help pages on deep research in ChatGPT; Claude support pages on using Research; Gemini Apps Help on Deep Research; Perplexity pricing page.