How Executive Search Firms Are Actually Using AI in 2026

Ask ten search firms whether they use AI and nine will say yes. The more useful question is what for, specifically, and how much of the output you can trust without checking it yourself.

The first half of that question has real data behind it now. Cluen’s most recent industry survey, reported by Hunt Scanlon in May 2026, found 73.5% of executive search firms report using ChatGPT, with Microsoft Copilot at 45% and Google Gemini at 12%. Only 9% of firms said they use no large language model at all, down from 22% the year before.

Whatever AI in retained search was eighteen months ago, it is not a pilot project anymore. It is infrastructure.

What firms are actually doing with it

The same survey breaks adoption down by task, and the pattern is consistent: AI is doing drafting and first-pass synthesis, not judgment.

• Candidate write-ups: 73% of firms use AI to draft these
• Role and job description writing: 69%
• Business development research: 55%
• Progress reporting and interview prep: 34%+

AESC’s own guidance to member firms lines up with that breakdown: transcribing and summarizing meetings, drafting position descriptions and correspondence, analyzing data sets, and automating the admin that used to eat a researcher’s afternoon.

AESC calls executive search “both an art and a science,” and in practice the line holds. AI does the first draft and the grunt work; a person still decides who is right for the room.

Notice what is not on either list. Nobody is reporting AI as the source of the market intelligence itself: the sized pool, the comp band, the real shortlist of who is out there. That is not because a model cannot help with that work. It is because doing it well requires something most firms’ AI workflow currently skips, which is sourcing.

Where it breaks: an unsourced AI deliverable is a liability, not a shortcut

In October 2025, Deloitte Australia had to refund part of a A$440,000 (about US$290,000) government contract after a report it delivered to the Department of Employment and Workplace Relations turned out to contain AI-fabricated content: a quote wrongly attributed to a federal court judge, and citations to academic papers and a book that do not exist.

Fortune covered the fallout after a University of Sydney researcher, reading the report, “instantaneously knew it was either hallucinated by AI or the world’s best kept secret” and found more than a dozen fabricated references. Deloitte reissued the report with the fabrications stripped out and a disclosure that generative AI had been used to help produce it.

That is a management consultancy’s report, not a search firm’s market map, but the failure mode is identical to typing “what’s the comp band for a VP of Operations at a $200M industrials company” into ChatGPT and presenting the answer as researched.

A language model will give you a confident, well-formatted number on request. It has no obligation to tell you where that number came from, and given a specific-enough question, it will fabricate a source before it says “I don’t know.”

For a market map going into a client pitch, that is the one failure mode a firm cannot absorb. A client who catches one made-up figure stops trusting every figure on the page, including the ones you got right.

What a properly sourced version looks like

That is the gap between “AI helped write this” and “AI helped source this,” and it is closable. A grounded market map does not ask a model to know the comp band for a role.

It pulls two or three comparable public companies’ DEF 14A proxy filings from SEC EDGAR for real, filed compensation numbers.

Then it anchors the floor against the Bureau of Labor Statistics’ Occupational Employment and Wage Statistics for that occupation and metro, and states plainly which number is a filed fact and which is an estimate built by adjusting for a private, earlier-stage company.

Market sizing works the same way. Census Bureau County Business Patterns gives a real establishment count for an industry and geography, a defensible denominator instead of “the market is large and growing.”

The AI’s job in that pipeline is what the Cluen survey shows firms already trust it for: drafting, structuring, and summarizing. The sourcing has to happen first, against primary data, or there is nothing trustworthy underneath the synthesis.

It is the same discipline behind every market map we build: each claim held to one of three tiers (fact, estimate, or inference), labeled as such, with every fact traceable to a link a client can check themselves.

For the full method, including how to scope and size the market before you profile anyone, see how to build a market map for an executive search.

It is also the same evidence a sourced map gives you in a pitch: see how to win an executive search mandate for how a sized pool and a filed comp read change a competitive decision.

Three checks before you trust an AI-assisted research output

Ask where a number came from

A link you can click and re-check, not a confident sentence. If a tool cannot show its source, treat the number as a guess.

Keep facts and estimates visibly separate

A wage floor from BLS is a fact. Adjusting it down for a private, early-stage company is an estimate. Label it that way and the one guess in a report does not discount everything around it.

Re-fetch before anything client-facing ships

AI drafts fast. A source that moved, got corrected, or never said what the model claims it did is a five-minute check away, and it is cheaper than a client catching it first.

That is the bar we hold every MergeSearch Market Map to: every comp figure, pool count, and market size traceable to a source the moment we hand it over.

If you want to see one built against a live mandate, start your free month: two Maps, on briefs of your choice, built during a free first month before you commit to anything.

Sources

• Cluen, most recent Top Trends in Executive Search Technology survey, as reported by Hunt Scanlon Media (May 2026).
• AESC, "Leveraging Artificial Intelligence for Executive Research Efficiency."
• Fortune, "Deloitte was caught using AI in a $290,000 report after a researcher flagged hallucinations" (October 7, 2025).
• U.S. Securities and Exchange Commission, EDGAR full-text filing search (proxy-statement compensation).
• U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics (wage distributions by occupation and metro).
• U.S. Census Bureau, County Business Patterns (establishment counts by industry and geography).