Where AI Tools Find the Experts They Recommend

Part 2 of Found, Seen + Remembered

Outsourcing begins with an admission that the work is bigger than the team. Ours was a project we did not have the capacity to build ourselves.

I asked a member of the team to find us an outside agency, and she started where anyone would. She looked up local agencies, searched the web and checked which agencies other brands had used for work people remembered.

We landed on three to interview and ran each one through a concept pitch. A small agency with some buzz around its work won the job.

Hiring someone for work you cannot do yourself means trusting what you can find about them. If you have ever hired a contractor or an agency, you have run that search. On the flip side, your next client runs a similar search on you and your business, and ChatGPT, Gemini, Claude and Perplexity do the first pass, answering with a short list of names before anyone opens a website. Which names appear depends on where those tools look, and this guide follows them there. The places that talk about you now shape the shortlist as much as your own site does.

For more on the new alphabet of search, what SEO, AEO and GEO mean and why they matter, read Part 1 of the Found, Seen + Remembered series.

The sources AI answers draw from

Five kinds of source appear again and again in the research on AI citations.

1. Community discussions such as Reddit, where participants answer each other’s questions.
2. Reference sites such as Wikipedia.
3. Video on YouTube.
4. Professional posts on LinkedIn.
5. News and editorial coverage from publications and trade press.

Our shortlist came from a mix of places: local listings, a general web search and the agencies other brands had trusted. AI tools build theirs the same way, from several kinds of source at once.

The mix changes quickly. In 5W’s State of AI Citations 2026, Semrush data across 230,000 prompts tracked what ChatGPT cited through August and September 2025.

Perplexity and Google AI Mode did not shift in the same way, and publications such as Forbes and Medium gained share as the two fell.

LinkedIn and AI recommendations

Of the five sources, LinkedIn is the one where you control the author name, the headline and the posting rhythm.

Someone deciding whether to hire you looks for the person behind the work. Semrush data reported by 5W puts LinkedIn in about 15% of Google AI Mode responses, and Gemini leans on LinkedIn alongside Medium, Quora, Reddit and Wikipedia.

Five habits build visible expertise there:

• A headline that names your expertise
• Long-form articles on your key topics
• Named frameworks you reuse
• Specific examples from your work
• A steady weekly cadence

A blog post adapts well into a LinkedIn article. Keep the answer-first opening, add your point of view to the first paragraph, trim the piece to one idea and link back to the full post.

Long-form video and AI citations

AI tools read video as text, so the transcript, the chapters and the description carry the weight.

A long video shows the work, which makes it the closest thing an expert has to a pitch that keeps running after the meeting ends.

OtterlyAI analysed more than 100 million AI citations in its YouTube Citation Study, and popularity carried little weight: 40.8% of the cited videos had fewer than 1,000 views. Long-form video took 94% of the YouTube citations, and description length was the strongest positive correlate. Most citations came from Perplexity and Google’s AI surfaces, with ChatGPT at 4.4% and Gemini at 0.2%.

Ahrefs measured something different. In its study of 75,000 brands, YouTube mentions correlated at about 0.737 with AI visibility across ChatGPT, Google AI Mode and AI Overviews. Correlation does not show cause, and the two studies together make video one source among five rather than a guarantee on any single tool.

Four elements give an AI tool text to read and quote:

Short clips extend your reach and lead viewers to the long video, where the transcript and the full explanation live.

Earned mentions and AI visibility

Mentions on sites you do not own are a source of their own.

The agency we chose had some buzz around its work before it ever pitched us. That is what earned mentions look like from the buyer’s side: other people were already talking.

Then they pitched, and it felt like they understood the assignment. They had taken the time to research our business before they walked in. They came with ideas and were open to reshaping their pitch around ours where that would make the project stronger. Everything our research had turned up held up in the room.

Ahrefs studied 75,000 brands and found that branded web mentions correlated with Google AI Overview visibility at 0.664, against 0.218 for backlinks. Brands in the top quarter for web mentions averaged 169 AI Overview mentions, more than ten times the 14 averaged by the next quarter down. Ahrefs notes that the data shows correlation only, across established websites with a domain rating above 40.

Ahrefs describes the pattern as one of language. Language models draw on how often a brand name appears beside particular topics, so mentions on sites you do not own build the link between your name and your topic, and that link is what an AI tool draws on when it recommends you.

Five ways to earn mentions without paid placements:

To pitch yourself as an expert source, choose one topic you can prove, write a two-line summary of your view, send it to someone who covers the topic and follow up once with something useful.

How each AI tool chooses sources

The same question can return different sources in each tool, so each one needs its own check.

5W, a communications firm, synthesised the following patterns from datasets published by Semrush, Seer Interactive and Profound.

Seer Interactive found that 87% of SearchGPT citations matched Bing’s top 10 results, against 56% for Google’s, in a sample of more than 500 citations. 5W estimates that only about 11% of domains are cited by both ChatGPT and Perplexity.

Ahrefs also found that ChatGPT mentions track traditional authority signals less closely than Google’s AI surfaces do, which may make ChatGPT a more open entry point for a newer brand. ChatGPT’s use of Bing makes it worth reviewing how your site appears in Bing as well as in Google.

Where to start

1. Write the five questions your clients ask before they hire you.
2. Ask each one in ChatGPT, Gemini, Claude and Perplexity, and record whether you appear and which sources are named.
3. Update your LinkedIn profile with a headline that names your expertise, an About section that names your method and links to your best pieces.
4. Publish one long-form video from a piece you have already written, with a transcript, chapters and a detailed description.
5. Pitch yourself to one podcast host or journalist who covers your topic.
6. Repeat the check each month, because results shift.

The agency that won our job was small, and it had buzz around its work. It won the way a lot of work is won: a name people had heard, backed by a pitch that showed they understood the assignment.

Whoever hires you, for one project or for the long term, will run a version of that search first. An AI tool now does the opening round, reading the discussions, reference pages, videos, professional profiles and news coverage that mention you, and it writes a short list. The pitch comes after, and what they find first decides whether you are invited to make it.

Found, Seen + Remembered builds in order. Part 1 laid the foundation: the new alphabet of search and the basics every page needs. This guide followed the names to where AI tools find them. Part 3, How to Write a Piece AI Tools Can Quote, comes next and turns to your own pages, so that every source above has something clear to cite. Start with the five questions, ask them in four tools and read what comes back. It is the same first step our team member took, from the other side of the table.

Sources

1. Ahrefs: An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied) 2. Ahrefs: Top Brand Visibility Factors in ChatGPT, AI Mode and AI Overviews (75k Brands Studied) 3. OtterlyAI: The YouTube Citation Study 2026 4. 5W: The State of AI Citations 2026 Last reviewed October 2026.