Get cited by ChatGPT: a practical AEO guide

Buyers already ask ChatGPT, Gemini, and Google AI Overviews their real questions. "Which tool for…", "Who builds…", "How do I choose…". If your site isn't a clear, crawlable source, the answer names someone else. Answer Engine Optimization (AEO) is the set of practices that raise your odds of showing up there.

This guide is for founders, marketers, and content owners who want their brand eligible and relevant for AI citations. No magic jargon: bot crawl, answer-first pages, dated proof, off-site mentions, and a simple measurement protocol.

Why citations matter (not only classic SEO)

Classic SEO still matters. It no longer answers alone "who does ChatGPT recommend on my topic?". Buyers type conversational prompts. Models rewrite the query, search the web (or partner indexes), then cite a subset of pages.

Google says it plainly in AI features and your website: no extra technical requirements for AI Overviews or AI Mode beyond being indexed and snippet-eligible. The rest is solid SEO, plus citations, mentions, and content that answers first, then expands.

The goal stays practical. For the questions your buyers actually ask, you want a clear, fresh, provable page that AI crawlers can fetch.

AEO, GEO, SEO: plain vocabulary

In practice AEO and GEO often chase the same outcome: being cited or mentioned in an AI answer. Jargon shifts by vendor. On X and in guides, the split is mostly marketing. What matters: citable pages + bot crawl + proof + off-site mentions.

The GEO paper by Aggarwal et al. reports up to 40% visibility gains in their lab experiments (GEO-bench). Read that as a lab signal, not a client promise. Citations, quotations, and sourced statistics help.

How ChatGPT and Google actually pick sources

Google (AI Overviews / AI Mode)

Google notes there is no secret AI-only markup. Best practices stay SEO basics: crawl, internal links, visible text, structured data aligned with visible text. AI Overviews may also use query fan-out: several related searches across subtopics. A citation does not require ranking #1 on the original query.

In Search Console, the Search generative AI control (global rollout noted late August 2026) lets you Include or Exclude AI Overviews, AI Mode, and generative Discover. Default: include. This is not the same lever as Google-Extended (model training). Mixing them up in copy or robots.txt is a common mistake.

ChatGPT (web search)

OpenAI states in Searching the web with ChatGPT that citations can be incomplete, outdated, or incorrect. Always check the source. To be eligible, allow OAI-Searchbot and traffic from the IPs in searchbot.json. No platform guarantees a citation.

Technical checklist: stay crawlable by AI bots

1. Allow OAI-Searchbot (and confirm CDN / WAF allow OpenAI search IPs).
2. Keep Search generative AI on Include, unless you have a strong reason to exclude.
3. Do not confuse Google-Extended (training) with the Search generative AI control (appearance in Search answers).
4. Avoid accidental noindex on pages you want cited (definitions, guides, FAQ, case studies).
5. Visible HTML text: key answers should not live only in images or scanned PDFs.
6. Sitemap and internal links to answer-first pages (see content section).
7. Watch logs: missing AI bots usually means infra, not only content.

An llms.txt file or invented schema does not replace these basics. OpenAI documents OAI-Searchbot crawl plus relevance, not placement guarantees.

Content checklist: which pages to make citable

A citable page answers in a few sentences, then expands. It dates facts. It links primary sources. It stays fresh (analyses relayed by Semrush often tie freshness to citations; treat that as an operating hypothesis, not a law).

Typical shape:

1. Short answer up top (definition or criteria).
2. Scannable table or list.
3. Dated proof (cases, sourced numbers, official links).
4. FAQ with real buyer questions.
5. Visible last-updated date.

Page types to prioritize

You don't need an "AEO site". Start with the formats models like to cite:

Category / offer pages. What you sell, for whom, with clear choice criteria. A sharp landing beats a cluttered homepage.

How-to guides. "How to do X", criteria, steps, pitfalls. Shape example: an article like how to choose an AI agency answers first, then goes deep.

Definitions and concepts. A page that clearly explains a technical term (e.g. Enterprise RAG or AI agents for business) becomes a natural source when someone asks "what is…".

Buyer FAQ. Real sales questions, not legal boilerplate. One question = short answer + expansion.

Case studies. Named outcomes, context, constraints. A case like LocService or Monka shows what shipped, not only what was promised.

Framework / compliance pages. If your market has rules (health, data, security), explain the framework with official sources. Don't invent badges. For health data in France, the HDS framework is covered in AI and health data: what HDS really means.

Proof and entity checklist (brand E-E-A-T)

Answer engines like clear entities. For a brand or product:

• Named cases or checkable proof with verifiable outcomes.
• Real institutional signals (funding, public programs, partnerships): only what is true and sourcable.
• Named authors on technical guides (not only "The team").
• Always-sourced stats; if the source is secondary, caveat or skip.
• About / team / method pages consistent with the articles.

Avoid invented certifications and self-awarded labels you don't hold. Explaining the framework and project experience is enough, and more credible.

Off-site checklist: become mentionable

Content alone isn't always enough. Practitioner playbooks (Hoffman, workflows like treg) stress:

1. Map the questions your buyers ask.
2. Publish answer-first pages.
3. Earn authentic mentions (forums, podcasts, trade press, GitHub, tech newsletters).
4. YouTube / conferences when the topic fits.
5. Quality citations (other sites naming you as a source).
6. Re-check mentions every 3 to 4 weeks.

Aim for the "core sources" of your sector (trade press, expert blogs, open-source docs). Skip viral percentages you haven't audited.

Measurement protocol: 5 to 10 buyer questions, every month

Without measurement, AEO is folklore. Simple protocol you can run yourself:

1. List 5 to 10 questions your buyers actually ask (conversational phrases, not Ads keywords).
2. Replay them on ChatGPT (with web search) and Gemini. Optional: Perplexity, Google AI Mode.
3. For each run note: citation (URL), mention (name without link), absence, tone (positive / neutral / negative), competitor cited.
4. Keep date, model / mode, and a screenshot.
5. Compare month over month. Tools such as Semrush AI Visibility, Mentions.so, or OSS kits (treg) are options, not requirements.

You measure eligibility and presence. No platform sells a guaranteed slot inside a model answer.

AEO myths to skip

1. "SEO is dead / AEO replaces it": Google still relies on the same fundamentals for AI Overviews.
2. "llms.txt guarantees a citation": not documented as a guarantee by OpenAI.
3. "Google #1 = cited by ChatGPT": only partial overlap; fan-out and model selection change the picture.
4. "AEO and GEO are opposites": often the same goal, different vocabulary.
5. "Blocking Google-Extended removes AI Overviews": no: training ≠ Search generative AI control.
6. "ChatGPT citations are always correct": OpenAI says otherwise.
7. Unaudited viral numbers (blog studies without method): skip or strong caveat. We omit them here.

Recap checklist (printable)

1. OAI-Searchbot allowed + search IPs OK.
2. Search generative AI = Include (unless deliberate exclude).
3. No Google-Extended / Search control confusion.
4. Answer-first pages on your buyer questions, visible text, real FAQ.
5. Dated proof + named authors.
6. Named cases or proof where true.
7. Regulatory framework explained without fake badges.
8. Internal links to offers, guides, and cases.
9. Authentic off-site mentions.
10. 5 to 10 prompts replayed monthly (ChatGPT + Gemini).
11. Refresh strategic pages within the year.
12. No promise of "we will get you cited".
13. Primary sources in your pages (official docs, papers, data).
14. CTA after value, not before.

FAQ

What is the difference between AEO and GEO?

AEO (Answer Engine Optimization) means practices to appear in AI answers. GEO (Generative Engine Optimization) mostly comes from the KDD 2024 paper and the same goal on generative engines. Treat them as one workstream: citable pages + crawl + mentions.

Do you need an llms.txt file?

OpenAI doesn't document it as a requirement to be cited. Priority: OAI-Searchbot, clear content, proof. llms.txt can be a discovery bonus, never a guarantee.

Can ChatGPT cite my site without a Google click?

Yes, via its web search and partners, if you're crawlable and judged relevant. That ranking isn't the classic Google top 10.

Are AI Overviews the same as ChatGPT?

No. AI Overviews / AI Mode are Google Search surfaces. ChatGPT is an OpenAI product with its own search. Levers overlap (good SEO, clear pages) but controls and bots differ.

Can anyone guarantee a citation?

No. You can aim for eligibility and relevance. No platform sells guaranteed placement inside a model answer.

Map your citable pages

The Vibe Company is an AI agency in Paris (YC W24). We build products, agents, and software factories that run in production. If you want to map which pages to make citable and your AI bot crawl, we can do that together.

Book a call about your AI visibility →