Executive Search Market Map: Build One With AI From Public Sources

A small executive search firm can build a market map with AI by describing the target in anonymised terms, listing companies and roles from public sources, scoring the longlist against a written rubric, and then verifying every name, title and date against a source it has opened. AI speeds the gathering and sorting. Verification stays with a person.

The warning that goes with this is simple. Language models state wrong things with great confidence, and an executive longlist is a document that clients and candidates both read closely. One outdated title or a mixed-up name in front of a client costs more credibility than the hours saved by a fast first pass.

Follow me on Instagram@sagnikteaches

What a market map is for

A market map is a structured picture of where the people who could do a role are likely to be. For an executive search, that usually means a list of target companies, the relevant roles at each, the people currently in those roles where they are public, and notes on why each is or is not a good prospect. It supports three decisions: whether the search is feasible, how wide to cast it, and whom to approach first.

Connect on LinkedInSagnik Bhattacharya

It is different from a candidate longlist, although one grows into the other. The map describes the territory. The longlist picks individuals from it. Keeping them separate stops a company-level judgement ("this firm is too small") from being mixed up with a person-level one.

Subscribe on YouTube@sagnikteaches

Keep the confidential brief off consumer tools

Settle this before any prompt is written. A live search brief often names the client, the reason the role is open, sometimes the incumbent who is leaving. Many of those facts are confidential to the client and sometimes to the market. Do not paste a named brief into a consumer chatbot.

The workable pattern is to describe the target in anonymised, generic terms in anything that goes to a general tool: "a finance director for a mid-sized industrial manufacturer with three sites and a recent acquisition". Keep the client's name, the reason for the vacancy and any named individuals in your own systems. If your firm has an approved business tool whose data terms you have read and whose training settings you control, you may be able to include more, but the firm's rules, not the tool's convenience, set the line.

Step 1: define the target on one page

Before searching, write a one-page target definition that anyone on the team could use. It should state the role in words a search engine would use, the likely company profile, the geography only as the client defined it, the size or revenue range, the sector and adjacent sectors, and the disqualifiers. Add what a good candidate has done, in three to five bullet points. An illustrative example:

ROLE: Finance Director (illustrative)
COMPANY PROFILE: manufacturers, 200-1,000 staff, 2-5 sites, private-equity or family owned
ADJACENT SECTORS: engineered components, packaging, building products
DISQUALIFIERS: listed groups over 5,000 staff; companies in administration
GOOD CANDIDATE HAS: led a finance team of 10+; supported an acquisition; presented to a board
ROLE TITLES TO MAP: Finance Director, CFO, Group Financial Controller, VP Finance

This page does two jobs. It gives AI something precise to work from, and it gives you a standard to score against later.

Step 2: build the company list from public sources

This is where AI helps most. Ask it to propose companies that fit the profile, then check each one against a source. A general model can suggest names from its training, but those lists are incomplete and can be out of date, so treat every suggestion as a prompt for a search, not a finding.

Using the target definition below, suggest 40 companies that plausibly fit. For each, give: company name, what it makes or does, approximate size band, and why it fits. Mark anything you are unsure of as UNSURE. Do not include any contact details, and do not name individuals.

TARGET DEFINITION:
[paste the anonymised page]

Then fill the gaps and verify with public sources a person can open:

• Company websites: about pages, leadership pages, careers pages, news and press releases.
• Annual reports and official filings, where companies are required to publish them, which show directors and sometimes senior executives.
• Trade association member lists and exhibitor lists for relevant events.
• Trade press and published interviews.
• Conference programmes, which show who speaks and in what role.

A model that can browse may help gather these, but it must still show you where each fact came from, and you open the source. A model without browsing is a brainstorming aid, nothing more.

A note on LinkedIn. Its help pages say that third-party software, including crawlers, bots and browser extensions that scrape the site or automate activity on it, violates its User Agreement and can lead to restricted or closed accounts. So do not point a scraper or an automation add-on at it. Look people up by hand within the terms you have accepted, and record what you found.

Step 3: map roles and people, with sources

For each company that survives, record the roles that match your title list and, where public, the person currently in them. Capture the source beside every entry. A spreadsheet works well, with one row per role and these columns:

The two columns that earn their keep are "source and date" and "status". Together they let you filter the sheet to verified rows before anything leaves the building.

Step 4: score the longlist against a written rubric

Once the map holds verified rows, a scoring rubric ranks them. Write the rubric yourself from the target definition so the weights reflect the client's priorities, not the model's. A simple version scores each person or company out of ten on a handful of criteria.

Say a candidate scores 8 on sector, 7 on scale, 9 on role, 6 on evidence and 5 on availability. The weighted total is (8 x 0.25) + (7 x 0.20) + (9 x 0.25) + (6 x 0.20) + (5 x 0.10) = 2.0 + 1.4 + 2.25 + 1.2 + 0.5 = 7.35 out of 10. Compare it against your threshold, for example 7.0 for the first shortlist, and note that any score below the evidence line leaves the person unconfirmed.

You can ask a model to propose the scores from the verified notes, but require a one-line reason for each score and a column for the consultant to override. Treat the model's scores as a first sort. The consultant's scores are the ones that count. A fair practice is to score on the work evidence and leave out personal characteristics. The tutorial on bias checks for AI screening explains why.

Step 5: verify, because this is where errors enter

Verification is a separate pass from collection, ideally by a second person. For every row that will appear in front of a client:

1. Open the cited source and confirm the name, the title and the company, matching spelling.
2. Check the date. If the page is more than a few months old, look for a newer one: a press release about a departure or an appointment.
3. Look for a second source for any senior person you plan to approach.
4. Confirm that the person is distinct from others with the same or a similar name.
5. Mark the row Verified, and record who and when.
6. Drop or mark UNVERIFIED anything you cannot confirm. Never smooth over a gap.

A realistic example of what this catches. The model's map lists a "Group Finance Director" at one manufacturer. The company's own leadership page, opened that morning, shows a different name and a new title. A trade press item two months old reports that the previous holder moved on. Without the check, a first call would have gone to someone who left, which is both embarrassing and avoidable.

Common ways AI maps go wrong

• Out-of-date titles. The model reports a role as it was when it learned it.
• Merged people. Two executives with similar names become one entry with a blended biography.
• Invented specifics. A plausible acquisition, qualification or previous employer appears that no source mentions.
• Company confusion. A subsidiary, a similarly named competitor or a parent company is mapped in place of the right one.
• Stale size bands. A company that has since been sold or merged still looks like a fit.

Each is caught the same way: a source, opened by a person, with a date.

A worked example with numbers

Say a four-person search firm has a mandate for a finance director at an industrial manufacturer, and wants a first map within two days. The consultant writes the target page, asks the model for 40 candidate companies, and removes eight on sight (wrong size, wrong sector) leaving 32. Searching public sources, the researcher confirms 27 still fit and finds that three are owned by the same group, so the pool is 25 distinct targets. At each, one to three roles match the title list: 41 role rows in all.

The researcher verifies each role against a company page or press item. Thirty-four are confirmed with a source and a date, five conflict (the title or the person differs between sources) and two cannot be confirmed and are marked UNVERIFIED. The consultant scores the 34 verified rows with the rubric, and 11 clear the threshold of 7.0. Those 11 form the longlist to discuss with the client. The remaining 23 stay on the map for later phases. Notice the funnel: 40 suggested, 25 real targets, 41 roles, 34 verified, 11 on the longlist. Every number above is illustrative, and the point is that each stage has a count you can audit.

Prompts for the steps after the company list

Turning source pages into role rows

Once you have opened a company's leadership page, paste the text and ask the model to structure it, which is a safer job than recalling facts from memory. The prompt below works on text you supply, so every output row traces back to something you can see.

From the page text below, list every person whose title matches any of: Finance Director, CFO, Group Financial Controller, VP Finance. For each give: name exactly as written, title exactly as written, and the sentence it came from. If no one matches, write NONE. Do not add anyone from your own knowledge.

A realistic output, with an item you would fix:

1. [name as written] | Chief Financial Officer | "[first name] joined as Chief Financial Officer in the spring..."
2. [name as written] | Finance Director, Components Division | "...leads finance for the components division."

The second line is a division-level role. Whether it counts against your title list is a judgement. Mark it for the consultant instead of letting the model decide.

Spotting the companies that look right and are wrong

Ask a second question over the verified map: which of these companies might be unsuitable for reasons the profile did not capture? The model may flag ones that are subsidiaries of a larger group, recently sold, or in a regulated niche. Each flag is a lead to check, not a fact.

Adjacent-sector mapping: where AI helps most

The strongest use of a model in mapping is lateral thinking. Given the target page, it can propose adjacent sectors where the same skills appear, such as packaging for a components search, or distributors for a manufacturing search. A consultant can then decide which adjacent sectors the client would accept. Do this early, because clients often say "same sector only" until they see that the pool of verified candidates is thin.

Take the illustrative finance director search again. A sector-only map gives 25 targets and 11 longlisted people. Adding two adjacent sectors, each of which the client agrees to, might add 15 targets and four more people above the threshold. You can present the choice to the client with the numbers beside it: staying in sector gives 11 people, widening gives 15. That is an easier conversation than a general plea for flexibility.

A short before and after for a leadership page

Take two versions of an entry on the same illustrative map.

BEFORE (from the model's memory): "[full name the model supplied], CFO at the company since 2021, previously led finance at a packaging group."

AFTER (verified): "[name as published], Chief Financial Officer. Source: company leadership page, opened [date]. Previous employer: not stated on the page. Status: Verified (title and company); previous role UNVERIFIED."

The before version contains three claims with no source. The after version contains one verified claim and one clearly marked gap. If the consultant wants the previous-employer fact, they find a source or leave it out. Never carry an unverified biography into a client document just because it reads well.

Confidential searches in practice

Some searches are confidential on both sides: the client does not want it known that the role is open, or that the incumbent is leaving. For those, keep a strict split. Anything that identifies the client or the vacancy stays in the firm's own systems. The map can still be built from the anonymised target page, and the link between the map and the client lives only in your private records. Name files with a code, not a client name, and restrict who can open the sheet. If an AI tool must see more than the anonymised page, it must be one the firm has approved for confidential client material, and the decision should be written down and made by a partner.

What to show the client, and what to keep back

The client sees the verified map and the shortlist, with sources available on request. They do not see UNVERIFIED rows presented as fact, the model's raw output or your working notes on people. Describe candidates by public career facts, not by guesses about their circumstances. And avoid any claim about a person's willingness to move unless it is public or they have told you.

Handling people's data properly

A market map is a collection of personal data about named executives, even when every item is public. Keep it in your firm's systems, restrict who can open it, and decide how long you retain it and when stale rows are deleted. Where data-protection law applies to you, take advice on your lawful basis for processing and on how you inform people. Do not paste the map into a general chatbot to "tidy it up". If you need AI help with a file that contains names, use a tool that your firm has approved for that kind of data.

Using the map after the search

A verified map is a firm asset. It shortens the next search in the same sector, and it supports business development: a client in the sector may value a one-page view of who is where. See AI for recruitment business development for how research feeds the pitch, and how consultants use AI to research clients before discovery calls for the shorter version of the same discipline.

Date every map, and set a review interval. Executive roles change frequently, so a map older than a few months should be re-verified before reuse, starting with the highest-scoring rows. And keep a short log of what the AI got wrong each time: the log tells you which prompts to tighten and which tasks to stop giving the model.

A first-week plan

1. Day 1: write the target page and the rubric, with the client's priorities in the weights.
2. Day 2: brainstorm companies with an anonymised prompt and cut the list by hand.
3. Days 3 and 4: collect roles and names from public sources, one row per role, with source and date.
4. Day 5: a second person verifies, the consultant scores, and the firm decides what goes to the client.

If you would like help setting up the sheet, the prompts and the verification step for your own firm, that is what my AI implementation consultation covers.

Questions search consultants ask about AI mapping

Can I let AI scrape LinkedIn for the map?

No. LinkedIn's help pages say third-party tools that scrape the site or automate activity on it break its User Agreement and can lead to restricted accounts. Use public company sources, press releases and filings, and use LinkedIn by hand within its terms.

Is it safe to put a confidential search into a chatbot?

Not into a consumer tool. A confidential brief names a client, a vacancy or a succession plan. Keep those out of consumer chatbots, and use anonymised descriptions or an approved business tool whose data terms you have read.

How accurate are AI-produced executive lists?

Treat them as leads. Models mix up people with similar names, report out-of-date titles and invent plausible details. Every name, title and date on a longlist needs a source you have opened.

How many companies should a first map include?

Enough to cover the realistic pool, which for a niche senior role is often a few dozen companies. Start with the best-fitting ones and verify them fully before widening. A smaller verified map is worth more to a client than a large unchecked one.

Further reads

• How to Evaluate the AI Features in Your Recruitment Software — Judging the AI features in your recruitment software.
• How Recruiters Use AI to Write Candidate Summaries Clients Read — Turning a longlist into summaries clients read.
• AI Screening: Bolt It Onto Your ATS or Switch Systems? — Bolt-on or switch, for screening tools.
• AI CV Screening for Recruitment Agencies: Setup and Safeguards — Setup and safeguards for CV screening.
• How Much Time Does AI Save a Recruitment Agency per Role? — What time AI saves per role in a small agency.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: LinkedIn Help, 'Prohibited software and extensions', read 4 November 2026; general executive search practice. Verify every company, title and date against its original source.