How to get cited by ChatGPT and Perplexity: a 7-step method
- The gate is access: a page that answers HTTP 429 or blocks OAI-SearchBot or PerplexityBot cannot be cited, whatever it ranks on Google. We found our own host returning 429 to both crawlers.
- Sources beat your site: in one 25-answer audit, 4 of 5 cost answers quoted competitor prices word for word from public pricing pages, and the client, absent from all 18 third-party surfaces we checked, was named in 0 of 25 answers.
- Bing is the blind spot: OpenAI names third-party search providers such as Bing behind ChatGPT search, so a site indexed on Google and thin on Bing is partly invisible to it.
- Timing: fixes on pages and profiles the engines fetch live can show in answers within days to weeks; what a model remembers from training changes only when it is retrained.
A buyer types "best [your category] for [their situation]" into ChatGPT or Perplexity and gets three or four names with a reason each, and your brand is either in that short list or it does not exist for that buyer. To get cited by ChatGPT and Perplexity, make your pages fetchable by their crawlers and indexed where they search, publish plain answers they can quote, and put consistent facts about your brand on the third-party sources they read for your category. The qualifier matters: those levers move the retrieval layer, which updates within weeks, while what the model remembers from training moves only when the model is retrained.
This guide is the method we run by hand in every audit, written out as seven steps you can start today, followed by the source classes the engines actually quote, a table of how the five major engines differ, and the limits nobody selling "guaranteed AI placement" mentions.
How do ChatGPT and Perplexity decide which brands to cite?
ChatGPT and Perplexity build a recommendation from two layers: what the model absorbed during training, and what it retrieves from the web while answering. For comparative buying questions the live layer does most of the work, because the engine searches, fetches a handful of pages, and assembles its answer from what those pages say. The brands named are the brands those few pages record.
Both companies document the crawlers that feed this layer. OpenAI states that OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features", separate from GPTBot, which crawls for model training. Perplexity states that PerplexityBot "is designed to surface and link websites in search results on Perplexity" and that its own index feeds the answers. A page those bots cannot fetch is a page that cannot be cited.
Every step in this method targets that retrieval layer, because it is the one you can move this quarter: retrieval-fed answers change within days to weeks of a recrawl, while memory-fed answers wait for the next model.
How do you get cited by ChatGPT and Perplexity? The 7 steps
Run the seven steps in order, because the early ones cap the later ones: a blocked crawler makes every content change invisible, and a thin Bing index hides a perfect page from ChatGPT. Each step names where the work happens and what finished looks like, so a marketer and a developer can split the list between them this week.
- Open your site to AI crawlers. Ask your developer to request your pricing page, homepage and top guides with the user agents GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot, three times in quick succession each, and to check robots.txt, the firewall and any CDN bot protection for rules that block or throttle them. Done when every request returns HTTP 200 and robots.txt allows OAI-SearchBot and PerplexityBot explicitly. On our own site the host's CDN was returning 429 to both bots while human visitors saw a healthy page; the fix was turning that CDN off.
- Get indexed on Bing as well as Google. Search
site:yourdomain.comon bing.com and compare the count with your sitemap, then verify the site in Bing Webmaster Tools (it can import your Google Search Console properties), submit the sitemap, and consider IndexNow so new and changed URLs are pushed instantly. Done when the Bing count matches the sitemap and your pricing page appears for your brand name. - Write the answer into the first lines of each key page. Open each buyer-question page and put a self-contained 40 to 80 word answer directly under the heading, with prices, numbers and claims in plain page text rather than images or animated counters. Done when the page still answers the question with JavaScript disabled. Our own homepage once shipped animated stat counters that started at zero, so a crawler reading the raw HTML quoted "0 fixes" where humans saw the real figures.
- Match your headings to the questions buyers type. List your five most valuable buyer questions, search your site for a heading that matches each in the buyer's words, and retitle or publish sections so each question has a heading in that wording with the answer directly beneath it. Done when every question on the list resolves to one page and one heading.
- Put your facts on the sources the engines read. Search your brand on the two or three review sites and directories your buyers use, on any marketplace your product belongs in, and in the comparison roundups that rank for your best-of question, and create or correct each listing with the same name, description, category and current prices. Done when every surface carries identical facts and a dated update. The table in the next section shows which surface decides which kind of question.
- Make your identity identical everywhere. Approve one sentence that says what the company is, deploy it word for word on your site footer, Organization schema, profiles and listings, and file corrections wherever a third-party record disagrees. Done when asking ChatGPT "what is [brand]?" returns your own wording or a faithful paraphrase of it.
- Publish dated, factual comparison content and pitch the independent lists. Write the comparison your buyers ask for, disclose that you publish it, keep every price exact and dated, and send a short email to each independent roundup that ranks for the question asking to be included on the facts. Done when the comparison is live with a visible update date and each roundup has one pitch on record.
The whole list is profile, page and configuration work that costs people's time rather than budget, and steps 1, 2 and 5 usually explain most of a brand's absence while also being the fastest to finish.
Which sources do AI engines take their recommendations from?
AI engines assemble recommendations from a short list of readable sources: software directories and review platforms, app marketplaces, comparison articles, official vendor pages, and community threads. Control your presence on those surfaces and you influence the answers. Across the audit panels we run, the recommendations the engines produce trace to these five classes again and again, with the mix shifting by question type.
| Source class | What engines take from it | Question types it decides | Your move |
|---|---|---|---|
| Directories and review platforms | Ratings, review counts, category placement, descriptions | "best X", "is X legit" | Complete and date every profile |
| App marketplaces | Listings, ratings, install counts, platform fit | "best X for [platform]" | List where your product belongs |
| Comparison articles and roundups | Rankings, positioning, feature claims | "best X", "[leader] alternatives" | Publish your own disclosed comparison; pitch the independent ones |
| Official vendor pages | Prices, features, company facts, quoted verbatim | "how much does X cost", "what is X" | Plain-text prices with an update date |
| Community threads | Practitioner opinions, warnings, niche picks | Long-tail and trust questions | Answer under your own name, disclosed |
Official pages get quoted word for word, especially on price. In one 25-answer audit for an ecommerce software client, four of the five cost-question answers repeated specific competitor prices verbatim from their public pricing pages, tier names included. The brands with plainly published prices owned the cost conversation, and the brand whose prices sat behind a slider was priced by its competitors' pages instead.
Absence from the sources means absence from the answers. The same client had no record on any of the 18 third-party surfaces we checked, and was named in 0 of the 25 answers, while the category leader appeared in 20 and the runner-up in 18. The engines were reading everyone's records but theirs.
Engines also have no freshness instinct, and the roundups they read are often written by the vendors in them. In that same panel an engine kept recommending a vendor whose website no longer resolves, because directories and old roundups still carried the name, and another engine cautioned that nearly every "top alternatives" list is published by one of the vendors while still using those lists to answer. Both facts are openings: a category full of stale records rewards the one vendor whose sources are current, and engines demonstrably read disclosed vendor-written comparisons.
How do ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews differ?
The seven steps apply to every engine, and the engines differ in where they retrieve from, whether they show their sources, and which crawler you must allow. The table below holds only what each company documents or what independent tests have shown; where a company has not published the detail, the cell says so.
| Engine | Retrieves from | Shows sources? | Crawler to allow | What moves it fastest |
|---|---|---|---|---|
| ChatGPT (search) | Third-party search providers such as Microsoft Bing plus OpenAI's own crawl; an August 2025 Semrush-reported test found Google results used on some plans | Yes, links in search answers | OAI-SearchBot (GPTBot is training only) | Bing indexing and unblocked pages |
| Perplexity | Its own index via PerplexityBot, plus live page visits | Yes, by default | PerplexityBot | Fetchable, answer-first pages on the sources it already cites |
| Google AI Overviews and AI Mode | Google's index; a page must be indexed and snippet-eligible, with no extra requirements per Google | Yes, supporting links | Googlebot | Normal Google indexing and snippet eligibility |
| Gemini | Grounded in Google Search | Yes, when grounded | Googlebot (Google-Extended controls grounding) | Google indexing |
| Claude | Web search with direct citations; search provider not published | Yes | ClaudeBot, Claude-SearchBot | Checkable, clearly sourced pages |
Two practical consequences fall out of the table. ChatGPT's retrieval runs through indexes most marketing teams never check, so Bing verification belongs at the top of the list, and the Semrush-reported finding that Google results also appear means both indexes are gates. Perplexity, Google AI Overviews and Claude show their sources, so reading the citations on your own buyer questions is a free map of exactly which surfaces to fix first.
How we trace a wrong AI answer back to its source
Getting named is half the job, because engines also state facts about you with full confidence, and some of those facts are wrong. In a delivered MentionShare audit, ChatGPT told buyers a client's monthly prices were annual figures, so every quote it gave read about 25% too high, and public company databases named four different people as the same company's CEO. A visibility score shows neither problem; both answers looked healthy and were false.
Tracing is the part of the work a scanner cannot do, and it is the core of a MentionShare audit. Laurynas Leskauskas, who runs every MentionShare audit personally, puts 5 to 30 of your real buyer questions through ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews by hand, reads every answer, records who is named and what is claimed, and follows each wrong claim back to the page or record that fed it. Each finding becomes a fix card with one owner, an effort estimate, the steps, and a "done when" check, and where the fix is a correction to a third party, the correction email is already written.
The free version of that process is the seven steps above, and the MentionShare audit + fix plan adds the tracing and the ordering. A MentionShare Snapshot runs your five most important questions for 149 EUR, and a Full Audit covers 30 questions with 15+ prioritized fixes and seven monthly re-measures for 690 EUR, both with a 14-day money-back guarantee. The MentionShare sample report shows a complete audit, fix cards included, before you pay anything, and our methodology write-up explains each step of the audit.
What will not get you cited?
The recommendations are assembled from retrieved sources, and neither OpenAI nor Perplexity sells a slot in them, so any offer of guaranteed AI recommendations is overselling; the sources are what you can change. Keyword repetition does nothing either, because the engines quote the passage that answers the question, and a page that repeats the phrase without answering gives them nothing to lift.
Astroturfed community praise fails twice over. Communities recognise fake enthusiasm and remove it, and the practitioner phrasing engines echo comes from threads where real users answered real questions. Answer under your own name, disclose who you are, and let the record build.
How to get cited by ChatGPT and Perplexity, in short
Open your pages to the crawlers, get indexed on Bing as well as Google, write the answer into the first lines of each page under a heading in the buyer's words, put identical facts on the directories, marketplaces and roundups the engines read, and publish dated, disclosed comparison content. Then ask the engines your own buyer questions and read who they name and what they claim, because the answers are the only scoreboard this channel has. If you would rather a person ran that check across all five engines and handed you the fixes in order, that is what an AI-visibility audit is for. Everything in this guide sits inside the wider practice of generative engine optimization, and if the AI answers in your category already name competitors and miss you, the ordering above is where to start.