Home→Insights→How to Get Mentioned by ChatGPT, Claude, and Perplexity
SEO, GEO & AEO
How to Get Mentioned by ChatGPT, Claude, and Perplexity
Sukhpreet Kaur
Data & Hosting Specialist
· 29 min
AI search models cite content with original numbers, named authors, and defended positions, structured to be quoted. Generic explainers get summarized away.
SEO, GEO & AEO Solutions
Looking for a seo, geo & aeo partner?
We build domain-led systems tailored to your industry and workflow. 12 years. 2,100+ engagements.
If you want to be mentioned by ChatGPT, Claude, or Perplexity, the honest first step is to understand what the model is doing when it picks a source. It is not running a Google ranking algorithm. It is reading a set of candidate pages, deciding which ones say something the training set does not already cover, and quoting from the ones that do. The page that gets mentioned is the one the model needs to ground its answer on. The page that gets summarized away is the one the model already knows how to write without you.
We run a production RAG-grounded chatbot on our own site, with a curated knowledge base of 60+ sources and guardrails on every output. Building and operating that system gave us a clear view from the other side of the mention question. We have watched a model decide which page in front of it is worth quoting and which is worth ignoring. The patterns are not mysterious. They are knowable, and they are the same across ChatGPT, Claude, Perplexity, and Google AI Overviews because every one of them is making the same underlying decision: is there something on this page I cannot reproduce on my own.
Below are the 5 patterns AI search models use to pick what to cite, the 3 kinds of content that never make the cut, the 5 questions to ask before you publish for mention, and the practical path from your content to a quoted source in an AI answer.
5
Patterns AI search models use to pick which sources to cite, the same across every major engine.
3-5
Sources a typical AI answer cites, so being one of them is a finite, contestable spot.
1.5B+
Weekly ChatGPT queries searching for sources to quote and cite back to the reader.
1
Test that decides mention: can the model quote your page without losing the meaning?
You will see the same patterns play out across ChatGPT, Claude, Perplexity, and Google AI Overviews, the same content that fails to be mentioned in all of them, and the practical work that moves a page from summarized-away to quoted-back.
The Same Query, Two Different Answers
The cleanest way to see how mention actually works is to ask ChatGPT or Perplexity a real question and look at what shows up in the source list. Same query, different answers, because the model has 2 different sets of candidate pages to read. The pages that win mentions look one way. The pages that lose them look another. Here is what each one sounds like.
Same Query, Two Answers
A Buyer Asks an AI: "How do I reduce no-show rates at my dental clinic?"
The model finds 2 candidate sources. Both cover the topic. Only 1 gets mentioned. The pattern below is the whole game.
Page A, The Recycled Explainer
Page title. "10 Ways to Reduce No-Show Rates at Your Dental Practice."
Content. Generic advice every clinic blog already published. Confirm appointments, send reminders, charge a deposit, communicate clearly.
Author. "By the Practice Marketing Team."
What the model does. Reads the page, recognizes every claim from a thousand other sources, paraphrases the obvious advice without citing the page, moves on.
Result: summarized away. No mention. No traffic. The page exists; the answer layer does not need it.
Page B, The Cite-Worthy Source
Page title. "Our 6-Month Test: A 19% Drop in No-Shows From Changing One Reminder."
Content. A specific test the clinic ran, with the dates, the patient segment, the exact reminder change, and the measured drop in no-shows. A chart. A method note.
Author. "By Dr. Priya Sharma, founder of a dental practice in Pune, with 8 years of operations data."
What the model does. Reads the page, finds an original number it has never seen before, quotes the 19% figure and the reminder-change detail, links the source.
Result: mentioned as the primary source. The page becomes the canonical reference for the question for as long as the data holds.
Both Pages Are Real. Only One Has Anything to Cite.
Page A and Page B both look like content marketing. The difference is what is in the cells. Page A reuses the same 10 reminders every blog already wrote. Page B carries one specific number from one specific test and a named person behind it. That is the whole gap between summarized-away and mentioned.
You can run this comparison on any topic your business writes about. Pull up the top 5 results in ChatGPT or Perplexity. Look at which ones got mentioned as sources. The pages that win all share the pattern of Page B. The pages that lose all share the pattern of Page A. The model is doing the same thing every time. Build for Page B and you become the source.
3 Kinds of Content That Never Get Mentioned, No Matter How Well Written
Before the playbook for what works, the cleanest filter is to know what does not. These 3 categories of content can be perfectly written and still never earn a mention, because the substance they carry is the substance the model already has. Spending content budget on them in 2026 is spending it where the answer layer cannot use it.
What Never Gets Mentioned
3 Kinds of Content the Answer Layer Skips Every Time
Each category fails the same way: it offers nothing the model could not already write from training. The cells below show why each fails and what it tends to look like.
Kind A
Generic "N Ways to Do X" Explainers
Every claim already appears in 50 other places. The model has seen the same 10 tips a thousand times during training, so the page adds nothing to cite. Perfect grammar, zero novelty, no quote-worthy line.
Tell: every bullet is something an experienced person in the field would already know.
Kind B
Recycled Summaries of Other Sources
A roundup of what other people have said about a topic. The model already absorbed all those sources directly during training, so it does not need a summary of them. The page is structurally between the model and its own knowledge, blocking nothing, adding nothing.
Tell: every paragraph cites or paraphrases another source rather than producing its own claim.
Kind C
Balanced Both-Sides Reviews
Content that carefully presents every view, hedges every claim, and recommends nothing in particular. The model does not need this either, because hedged consensus is exactly what the training set is full of. Sources get mentioned when they commit; balanced reviews give nothing to commit to.
Tell: the page never says "we recommend" or "in our experience" and refuses to take any specific position.
Each Kind Fails the Same Way
Generic explainers, recycled summaries, and balanced reviews all share the same flaw: they offer the model nothing the model does not already have. Mention is the model saying "this page added something I needed." If your page does not, nothing about how well it is written can change the decision.
The filter is brutal but useful. Before drafting any new piece for AI search, ask which of the 3 kinds above it might fall into if you write it the obvious way. If the honest answer is "one of these," the brief itself needs to change. The fix is never better writing on top of the same substance. The fix is different substance underneath.
5 Patterns AI Search Models Reward When Picking Sources
Across ChatGPT, Claude, Perplexity, and Google AI Overviews, the patterns that lift a page from candidate to named source are consistent. Build for these and you compound across every major engine at once.
A Number or Claim the Model Has Never Seen
The single strongest pattern. An original number from your operations, a finding from a test you ran, a specific result from a project, a piece of internal data. The model will quote the number and link the page because it cannot generate the figure on its own. One specific data point on a page often outweighs 2,000 words of generic surrounding context. We have watched this play out repeatedly on our own RAG-grounded chatbot: the chunks that get retrieved are the ones with specific figures.
A Quotable Sentence in a Chunk the Model Can Lift Cleanly
AI models retrieve passages, not pages. They look for a sentence or 2 they can extract and present back to the user without rewriting heavily. Short paragraphs with one specific claim, immediately followed by the supporting reasoning, retrieve well. Long paragraphs that bury the claim 8 sentences in get compressed away before quoting. The chunk structure decides whether the model picks your sentence or someone else's.
A Named Author With a Real Reason to Be Writing This
A byline with a real name, a real role, and a real bio is a mention attribution handle the model can attribute the claim to. "By the Marketing Team" is structurally invisible to attribution. "By a named person with a stated role in the business" is something the model can cite back to the reader. Author identity is increasingly weighted across every engine, because the engines are trying to defend against generative slop and identity is one of the few cheap signals that still works.
A Defended Position the Model Can Cite as a Stance
A page that says "in our experience, X works and Y does not, and here is why" gives the model something definitive to attribute. The model cannot generate a defended opinion from training; it can only paraphrase the average. A clear position with the reasoning visible turns your page into a source for the position itself, not just a paraphrase of the consensus. We see this consistently: opinionated content gets mentioned more than balanced content, even when the balanced piece is technically more complete.
A Track Record of Other Mentioned Pages on the Same Site
Sites that already get mentioned get mentioned more, because every engine builds an internal sense of which domains are trustworthy for which topics. The first mentioned piece is the hardest. Every one after it lifts the next one. This is why mention work compounds: ship 3 first-party-content pieces on the same topic cluster and the engine starts to treat the whole cluster as authoritative on that subject, which means even the medium-strong pieces start getting picked up.
None of the 5 requires being a famous brand. They require having something specific to say, saying it with a real name attached, and saying it in chunks the model can lift cleanly. Practices that hit all 5 get mentioned consistently. Practices that hit any 1 with discipline start getting mentioned within a month or 2.
3 Honest Cases Where Mention Is Not the Right Goal
Not every page on your site needs to win an AI mention. Knowing where the goal is something else keeps the mention budget focused on the pages where mention actually pays back. These are the cases where you should leave the page alone and not optimize it for AI search.
Transactional and Service Pages
Pricing, checkout, account settings, terms of service, refund policies. These exist to be found by a customer with intent to act, not to be mentioned by an AI summarizing a topic. Spend no mention budget on them. Make sure they load, parse, and convert, and let the rest of the site carry the mention work.
Routine Documentation and Help-Doc Walkthroughs
"How to reset your password," "where to find your invoice," "how to add a user." These need to work for your existing customers, not impress an AI search engine. The outcome is the customer solving their problem. Skip the mention patterns and write the clearest, shortest version of the answer that gets them unstuck.
Long-Tail Definition Pages and Glossary Entries
A definition of a common industry term that someone searched for. AI search will give the user the definition inside the answer itself most of the time, with no link click. A clean glossary page still earns some incidental traffic, but it is not a mention target. Ship it cleanly, move on, and put the careful work where the mention pays back.
The Forward Read
The bar for mention is going to keep rising, in 2 ways at once. First, the models will keep getting better at producing generic answers from training alone, which means the floor everyone shares (recycled explainers, balanced reviews, definition pages) will keep losing relevance. Second, the models will keep getting better at recognizing original substance when they see it, which means the lift from real first-party content will keep growing. The 2 lines diverge every quarter. Sites that start producing original numbers, real positions, and named-author content through 2026 will be the named sources of 2027, with a compounding lead. Sites that keep iterating on the same generic content will keep watching the answer layer write the answer to their topic without ever needing them.
5 Questions Before You Publish a Page for AI Search Mentions
Whether the brief came from marketing, the founder, or an SEO agency, these 5 questions decide whether the page can earn mention or will fall into the summarized-away pile. Ask them before the draft, not after.
Is There a Real Named Author Behind This?
A real human with a real role and a real reason to be writing this piece. "By the team" disqualifies a page from attribution. A named founder, a named clinician, a named engineer, a named operator with the bio that explains why they would know about this is the mention handle the model needs. Without it, the model has no one to attribute the claim to and will pick a competitor who does have a name on the page.
What Specific Input Goes Into This Piece?
A number you generated, a story you lived, a stance you defend, a finding from a test you ran, a dataset you own. If the honest answer is "research and a writer," the piece will look like every other recycled explainer in its category. The input is the predictor of the output. The model is reading for the input.
Does It Take a Position the Model Cannot Reproduce?
A clear stance defended with the reasoning visible on the page. Not "5 things to consider," not "it depends on your goals." A specific recommendation an expert is willing to defend. The model can paraphrase consensus content from training alone; it can only cite content that commits to a position. Commit, defend, and the page becomes a source for the position itself.
Is It Structured to Be Quoted in 1 to 2 Sentences?
A claim a model can lift in a single sentence, with the supporting reasoning right next to it. Short paragraphs. Clear sub-headings. A specific figure that lives in its own line rather than buried in a 200-word block. The model is looking for an extractable passage, not a long well-written essay. Sites that write in extractable chunks get mentioned far more than sites that write in long sustained prose, even when the prose is better.
Will the Mention Still Hold in 12 Months?
The original input behind the piece should still be distinctive a year from now. Numbers from a test that aged out, positions everyone has now adopted, stories tied to a fading product all lose mention weight fast. The pieces that compound are the ones whose first-party input stays distinctive for years. A 12-month test is the right horizon to think in, because that is roughly how long a piece needs to keep earning mentions to pay back the cost of writing it well.
From Your Content to a Mention in an AI Answer
The journey from a page on your site to a quoted source in an AI answer runs through a specific path. Understand each step and you can build for it. Skip a step and the page never gets into the candidate set, much less the mention list. Below is what the path actually looks like for every page you ship.
Content to Mention
The 5 Steps Between Your Page and a Mentioned Source in an AI Answer
Step 1
Write With an Original Input
A number, a story, a position, a finding. Without this, the page never makes it past Step 2 no matter how clean the technical work is.
▸
Step 2
Structure for Extractability
Short chunks, clear claims, a named author, structured data markup, an llms.txt entry. The technical wiring that lets the model find, parse, and attribute the page.
▸
Step 3
Get Crawled and Indexed
Search engine bots and AI search bots find the page, parse the chunks, and add them to the candidate retrieval pool. This step depends on technical infrastructure but is usually automatic once it is wired right.
▸
Step 4
Get Retrieved on Query
A user asks an AI engine a question your page answers. The engine retrieves a set of candidate chunks. Your page is in the set if Steps 1, 2, and 3 went right.
▸
Step 5
Get Picked as a Mention
From the candidate set, the engine picks the 3 to 5 sources whose chunks it actually quotes and cites. Step 1 decides this step. Without an original input, even retrieval does not become mention.
Step 1 Decides Step 5
Most sites that work on AI search mention spend most of their budget on Steps 2 and 3, the technical wiring. The wiring is necessary, never sufficient. What decides whether your page becomes a mention is Step 1: did you ship something with an original input the model could not reproduce from training. Get that right and every step downstream tends to follow.
The path is the same whether you are writing for ChatGPT, Claude, Perplexity, or Google AI Overviews. Build all 5 steps for one page and every engine that uses the same retrieval pattern picks you up at once. Build only the technical steps and you get found, parsed, and ignored, because there was nothing on the page worth citing in the first place.
Frequently Asked Questions
How do ChatGPT, Claude, and Perplexity actually pick which sources to cite?
Each one runs a retrieval-and-ranking step on every query: pull candidate pages that look relevant, rank them by signals about quality and originality, and pick the 3 to 5 strongest to cite back to the user. The retrieval is similar to a search index. The ranking is where the mention choice is made, and it heavily favors content that carries something the training set does not already cover: original numbers, first-hand stories, defended positions, named-author content. The mechanics differ slightly between engines, but the underlying decision is the same: cite the page that adds something, paraphrase past the page that does not. We have built and operated this kind of system on our own site, so the patterns are not theoretical.
How long does it take to start getting mentioned after we publish?
For a page that genuinely carries an original input and is structured to be extractable, the typical window is 2 to 8 weeks from publish to first mention. The variance is real and depends on how quickly your site gets crawled, how unique the input is, and how active the topic is in AI search queries. We see pages with strong original numbers get picked up in under a month consistently. Pages whose only differentiator is the technical wiring take much longer or never get there. The reverse pattern, technically clean but generic content, almost never starts getting mentioned at all. The work is front-loaded: get the input and the chunk structure right, then wait for the crawl and retrieval cycle to catch up.
Does the same content work across ChatGPT, Claude, and Perplexity?
Mostly yes. The retrieval and ranking patterns differ in detail across engines, but the underlying decision (cite content that adds something to the answer) is the same. A page that wins mentions from one engine usually wins them from the others within a few weeks of indexing. The exception is Perplexity, which leans more heavily on Reddit threads and forum content for certain query types, so for topics where Reddit dominates the conversation, a first-hand-experience page from your own site is the cleanest competitive move. ChatGPT and Claude both lean toward published, authored content with structured data, so the same pattern works for both. Build for the universal pattern (original input, extractable chunks, named author, structured wiring) and you compound across all 3 at once.
How important is the technical structured-data wiring vs the original content?
Both matter, but they are not equal. The technical wiring (structured data markup, llms.txt, author bylines, chunk structure) gets your page into the candidate set for retrieval. Without it, the engine may not see you. But the wiring on its own never wins mentions, because the mention decision is made on substance. A page with perfect wiring and generic content gets retrieved and ignored. A page with rough wiring and an original number still gets mentioned, because the engine finds the substance one way or another. The honest spend ratio is roughly 1 to 4: spend a quarter of the budget on the wiring once and 3 quarters on producing original content that justifies the wiring being there in the first place.
How do we measure whether AI search is citing our content?
Direct measurement is the cleanest path: query the major AI search engines on a set of target prompts on a recurring schedule and look for your domain in the mentions. This is bespoke work today, not a turnkey tool, but it can be automated. Indirect measurement uses referrer signals from ChatGPT, Claude, Perplexity, and Google AI Overviews where they leak through, watches for brand mentions in conversational queries, and tracks ranking on the questions AI answers most often. Most analytics tools see only a thin slice of AI search traffic today, so the synthetic check on target prompts is currently the most reliable signal. The measurement gap is real and shrinks every quarter, but the practical answer in 2026 is to instrument both paths and run them in parallel.
Should we still publish content that will not be mentioned?
Yes, for the right reasons. Transactional pages, help-doc content, product pages, and basic glossary entries serve real business outcomes that have nothing to do with mention. The mistake is publishing generic explainer content on topics your business cares about under the assumption it will earn mentions, when it will not. Run the filter before the brief: if the piece would be a generic "N ways to do X" explainer, a recycled summary, or a balanced both-sides review, the brief itself is the problem and no amount of technical optimization will fix it. Either reshape the brief to carry an original input or skip the topic and put the budget on a piece that can carry one.
Can Entexis help us build content that ChatGPT, Claude, and Perplexity will cite?
Yes, that is exactly the work we do. We run a production RAG-grounded chatbot on our own site with a curated knowledge base, so we have direct operating experience of which pages a retrieval-and-mention system picks up and which it ignores. We start with what your business actually has that is first-party: the operational numbers, the project stories, the defended positions, the named experts. We turn that into pages structured to be extractable, with the author bylines, the structured data, and the chunk shape the engines reward. We run a synthetic mention check on target prompts so you know what is showing up where, and we iterate based on what the engines actually pick up rather than guessing. The source layer stays yours, the technical layer stays in your stack, and the mentions compound. If your content has stalled while AI search rewrites everything around it, the answer is probably not more pages. It is the right input on the right chunk structure with the right name on it.
For the broader thesis behind this, why first-party data is the AI search moat across every engine and surface, the anchor piece is here: Why First-Party Data Is the AI Search Moat.
The most important thing to take from this is the reframe. Getting mentioned by ChatGPT, Claude, or Perplexity is not a technical SEO problem with a checklist solution. It is a content problem with a substance answer. The model is reading your page and asking one question: does this give me something I cannot already produce on my own. Answer yes and you become a source. Answer no and you remain a candidate that the engine retrieved, parsed, and skipped on its way to quoting someone else.
Want Content the Answer Layer Will Quote, Not Skip?
At Entexis, we build content that AI search engines cite, starting from the substance and working out to the wiring. We pull the original numbers from your operations, the real stories from your projects, the defended positions from your founders, the named-author bylines from your team. We structure every page for extractability, wire the technical layer (structured data, llms.txt, sitemap, author markup), and run synthetic mention checks across ChatGPT, Claude, Perplexity, and Google AI Overviews so you know what is showing up where. The content stays yours, the mentions compound, and the work is portable. If your content spend has stopped paying back since AI answers ate the search results page, the answer is probably not more pages. It is the layer underneath. Start the conversation with Entexis.
Ready to Win AI Search?
Manual SEO cannot keep pace with GEO and AEO. We build the workflows and automation that keep your brand visible across AI answer engines. Tell us what you need.
We'll get back within one business day.
Thank You!
We've received your message and will get back to you within one business day.
