Insights

How to get your site cited by ChatGPT, Perplexity and Gemini

You cannot make ChatGPT, Perplexity or Gemini cite your site, but you control what they need first: permission to fetch the page, HTML they can read without running your JavaScript, a company they can identify, and a sentence worth quoting. On audacian.eu those are robots.txt, prerendered pages, one Organization record and answer-first openings.

By Alan, founder of Audacian · 7 min read

Published {date}

What you control, and what you do not

A citation is the model’s decision. It depends on what the system was trained on, what it can fetch at answer time, and how it chooses sources, and none of that is a setting on your site. Anyone selling a guaranteed mention is guessing.

What you can do is remove the reasons a system would skip you: a crawler that is blocked, a page that is empty until scripts run, a company described three different ways, or a page whose opening is about you instead of the question. The rest of this article is that list, done on this site, with the files you can open yourself.

1. Let the crawlers in, by name

Start with robots.txt, because a disallow there ends the conversation before anything else matters. Ours allows everyone by default and then names the crawlers explicitly, so the policy is readable rather than implied: Googlebot, Bingbot, GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended. The file ends with the sitemap URL, and a comment points at llms.txt and llms-full.txt.

Each of those user agents means something specific to the company that runs it, and they are not interchangeable: some fetch pages for search features, some fetch a page because a person asked about it, and Google-Extended is a control for how Google may use your content in its AI products rather than a separate crawler. Read each owner’s documentation before you block or allow one, and write the decision down in the file. Allowing them is a policy choice. It is not evidence that any of them cite you.

2. Ship the text in the HTML

Not every crawler runs JavaScript, and the ones that do may not wait for it. audacian.eu is a React app, but every indexable page is prerendered at build time, so the HTML a crawler fetches already contains the headings, the paragraphs and the structured data. Interactive parts load afterwards without changing the text.

The same pages carry a canonical, a title, a meta description and hreflang for English, Spanish, Romanian and French with an x-default, so a system that fetches the Spanish page knows which English page it matches. The sitemap lists only indexable URLs, and its lastmod comes from the last change to each page’s source rather than the build date, so the date means something.

Production deploys ping IndexNow with the sitemap’s URLs. That tells the search engines that support it what changed; it does not decide whether they crawl it.

3. Publish llms.txt, generated, not hand-written

llms.txt is a proposed convention (llmstxt.org) for a short plain-text map of a site aimed at language models. We do not know which systems read it, and we do not claim any do. It costs little, and a hand-maintained copy is worse than none, so ours is generated at build time.

audacian.eu/llms.txt opens with one sentence about who we are, then lists our work grouped by service, built from the same proof registry the pages use, then the translated URLs of the main pages. audacian.eu/llms-full.txt lists every indexable URL with the exact title and meta description that page ships. Both come from the same route list as the sitemap, so a page that is removed from the site disappears from all three at once. Videojam generates its llms.txt from the same indexation gate as its sitemap, so pages kept out of the index are kept out of llms.txt too.

4. Be one entity, everywhere

A system that writes “according to Audacian” needs to be sure which Audacian. Our JSON-LD has one Organization with a stable @id on every page, the same legal and brand name, and sameAs links to our live profiles on LinkedIn, Crunchbase, Clutch and YouTube; we add a profile to sameAs only once it is live and matches our details. The founder is a Person linked to that organization, case studies are CreativeWorks created by it, and service pages point at it.

The same discipline applies to products. Mindscape, our IQ and EQ self-assessment, is described as a SoftwareApplication with its own @id and the disclaimer it uses on its own site: for entertainment and self-reflection, not a clinical or validated assessment. Searchless is described without offers, ratings or an install link while it is not in the Chrome Web Store. If the structured data claims more than the page does, you are teaching a system to distrust you.

Mindscape case studySearchless case studyCheck your own site with the AI search readiness checker

5. Write the sentence you want quoted

A citation lifts a passage. If the passage that answers the question is in the fourth paragraph, behind a story about the company, it is less likely to be the one lifted. Every service page on this site opens with a sentence that answers its own title, marked as the page’s answer in the HTML, and the evidence follows.

Envision Deco’s idea pages follow the same rule: the Japandi living room page says what that room is in its opening, then shows the palette, the floor plan and the products. Its Article, FAQPage and BreadcrumbList markup repeats the visible text and nothing else.

Give figures with their date and their source in the same sentence. “970 indexed pages” is a claim; “970 indexed pages in Videojam’s coverage export of 23 October 2026” is a citation someone can check.

SEO vs GEO vs AEO

6. Measure it as its own number

Search Console’s AI-features export is separate from web search. For Videojam, 1 August to 31 October 2026, it shows 749 impressions with no clicks and no queries, next to 12,486 impressions and 272 clicks on web search. 729 of those AI impressions fall from 23 September onward, the day after the crawl-path work, which is the same correlation the web search report shows. It is not a count of citations, and it says nothing about how often a model named Videojam.

Referrals from AI assistants show up in your analytics as referral traffic when they do happen. If they do not show up yet, that is the measurement. It is not a reason to estimate one.

The Videojam SEO case

The checklist

If you do one pass on your own site this week, do these, in this order:

  • Open robots.txt and confirm you are not blocking the crawlers you want. Name them explicitly.
  • Fetch a key page with JavaScript disabled. If the text is missing, fix rendering before anything else.
  • Check that every indexable page has a canonical, a title, a description and, if you are multilingual, hreflang.
  • Write your Organization once: one @id, one name, sameAs only for real profiles.
  • Generate llms.txt from your routes instead of writing it by hand.
  • Rewrite the opening two sentences of your five most important pages so they answer the page’s question.
  • Add FAQ markup only where the questions are printed on the page.
  • Report AI impressions and AI referrals separately from search, with dates.

Questions this article answers

Should I allow GPTBot and other AI crawlers?

It is a policy decision, and each crawler means something different to its owner. audacian.eu allows Googlebot, Bingbot, GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended, named in robots.txt. Blocking a crawler and then hoping to be cited by the product that uses it does not work.

Does llms.txt get my site cited?

Not on its own, and we do not know which systems read it. It is a cheap, plain-text map of the site. Generate it from your routes so it never disagrees with the sitemap.

What schema helps AI systems identify my company?

One Organization with a stable @id, used on every page, with sameAs links only to profiles that exist, and people and services linked back to it. Consistency matters more than the number of types.

Can you guarantee a mention in ChatGPT or Perplexity?

No. A mention is the model’s choice. We can show that the crawlers are allowed, the HTML contains the text, the entity is consistent and the page answers its question in the first sentences.

Tell us what you're launching.

Book a 20-minute intro call, or email a few lines about what you're launching. We'll reply with a rough plan and a price range, whether or not we end up working together.