What it takes to get found in ChatGPT and Claude

Ask ChatGPT which estate planning attorneys handle families with businesses in several states. You get a few names, a paragraph of reasoning about each, and a short list of source links underneath.

Two separate steps produced that answer. One chose the names. The other went looking for pages that supported them.

Most advice about getting found in chatbots treats those steps as one. The tactics that follow from that mistake are easy to sell and easy to do. That is part of why they spread.

Being cited and being named are two different results

Your page can be used as a source in an answer that recommends somebody else. Seer Interactive gave that outcome a name, ghost citations. Their February 2026 study covered 541,213 assistant responses across 20 brands and six platforms.

The pattern in their data is the useful part. A brand's pages were cited at roughly five times the rate in answers that also said its name. Answers that left the brand unnamed cited its pages far less often. If retrieval were choosing the names, those two rates would sit close together.

The reading that fits is a sequence. The model composes its answer first, naming the people and firms it associates with the question. A retrieval step then goes out and finds pages that support what it already said.

Seer marks this as behavioral evidence rather than observed architecture. Nobody outside the labs can watch the sequence directly. Treat it as the best current explanation.

What follows from it holds either way. Your website is mostly competing to clear the second step. Whether your name appears in the first one depends on what the model absorbed about you from everywhere else.

So a strong Google ranking and an absent name in ChatGPT can sit side by side without contradiction.

The first gate is being in the index each assistant reads

An assistant answering a live question does not browse the open web. It queries a search index, pulls a handful of pages, and writes from those. Google calls the technique grounding, or retrieval-augmented generation, in its guidance for site owners. The same page describes query fan-out: The model fires several related searches at once. It gathers results from all of them.

The index differs by assistant. Google's AI Overviews and AI Mode read Google's index. ChatGPT reads its own crawler's index alongside Bing. Claude's web search runs on Brave. These are commercial arrangements and they move, so treat the mapping as a snapshot.

The practical version takes an afternoon. Verify your site in Bing Webmaster Tools as well as Google Search Console. Plenty of sites that rank on Google were never submitted to Bing. Check that your robots.txt file is not blocking the crawlers that feed assistant search results. A page no index holds cannot be retrieved by anything downstream.

What other people publish about you does most of the work

Muck Rack's research team has run the same study three times. The May 2026 edition covered more than 25 million links cited by ChatGPT, Claude and Gemini.

Earned media supplied 84% of those citations, journalism alone supplied 27%, and paid or advertorial content supplied 0.3%. That is the ratio between what you publish about yourself and what other people publish about you. It is measured at the moment an assistant decides what to read. The earned media figure has held between 82% and 89% across all three editions since July 2025. A pattern that steady is unlikely to be a quirk of one model update.

Ahrefs reached the same shape from a different direction. Across 75,000 brands, branded web mentions correlated with assistant visibility at 0.66 to 0.71. Backlinks came in around 0.22, and the number of pages on a site around 0.19. Mentions inside YouTube titles, descriptions and transcripts correlated most strongly of anything tested, at about 0.74.

Read those numbers with their limits attached. Ahrefs filtered for established domains and says outright that correlation is not causation. The signals may be markers of brand strength rather than causes of it. What they rule out is the idea that publishing more pages or buying more links is the lever.

For an expert practice, the work this points to is ordinary and slow. Guest articles in the publications your buyers read. Podcast appearances where the show notes carry your name and your field. Association directories, conference programs, and being the person a journalist calls for a quote. Video, where the transcript names you.

One detail is worth the extra minute. A bio line saying who you are and what you do teaches the model an association. The model can recall that association later. A quote attributed to a bare name teaches it much less.

One record of who you are, repeated everywhere a machine reads

Machines resolve people the way a careful stranger does, by checking whether the facts agree. Your name, your field, your city, and your title should read the same everywhere. That means your site, your LinkedIn profile, your podcast bios, your directory listings, and your bylines.

Small mismatches split one strong record into several weak ones. A middle initial in one place and not another. A job title left over from two roles ago. A description that says strategy consultant here and management consultant there.

Wikipedia and Wikidata carry weight, and both run on strict notability rules their communities enforce. Pursue them when the published record supports it and leave them alone when it does not.

This is the part of the work Nexus handles through interviews. The recorded sessions produce one description of the person and the business, written down once. Everything afterward copies from that description.

Publish the answers only you can give

Google's guidance puts unique, non-commodity content above every other suggestion it makes. It draws the line at first-hand experience. A summary of what already exists restates what anyone could say. A first-hand account carries something the systems cannot get elsewhere.

For an expert, that means writing the answers you give in the room. The case that went sideways and what it taught you. The advice you give that contradicts the textbook. The question buyers keep asking in month two, answered at the depth you would answer it in person.

Two effects follow. Assistants get pages worth quoting. Readers get the experience density the search systems have rewarded since well before chatbots arrived. The work compounds in ordinary search at the same time.

What you can ignore

The same guidance page names several popular tactics as unnecessary for Google's systems.

• Machine-readable files written for AI, including llms.txt. Google says its systems do not use them. Independent monitoring of assistant crawlers finds the file is rarely requested.
• Breaking pages into small chunks for machines to digest.
• Rewriting your content in a special style for AI, which the systems do not require.
• Chasing mentions nobody meant, which spam systems are built to discount.
• Heavy investment in schema markup, which Google states is not required for generative AI search. Keep it as ordinary site hygiene.

Google also warns against tools claiming access to its internal ranking systems, since no third party has that access. The paid layer sold on top of this work deserves the same scrutiny.

What this work will not do

This work will not send you much traffic. Ahrefs measured that Google sends 190 times more traffic to websites than ChatGPT does. ChatGPT's click-through rate runs 96% below Google's. The value of being named sits elsewhere. It puts you on the shortlist a buyer carries into the next step.

It will not move quickly. Seer published a test on a client page cited more than 100 times with no brand mention. They added explicit brand language and saw no change 29 days later. Models absorb the web on their own schedule, so months is the sensible expectation.

It will not hold still. Ask the same question twice and the list of names can change. That makes any single check a weak measure, and a repeatable panel of prompts the better instrument.

And nobody can promise you a place in those answers.

The systems name the people the web already talks about. So the work that gets you named looks like reputation rather than optimization. That means consistent facts, other people's coverage, and answers only you could write.

Ask the question your best buyer would ask, in their words. Then read the sources listed under the answer. Notice how many are pages somebody else wrote about somebody else.