Writing pages AI assistants can cite
You get cited by AI when your page answers the question in the first lines, with a structure assistants can quote. This guide shows the exact pattern, what to avoid, how to let the crawlers in, and how to test it. It is for founders who ship their own site.
By Théophile Louvart, founder of Porteur · Updated 14 September 2026 · Markdown
The page pattern assistants cite
Write the page to answer a question in plain words, then show the facts they can quote. Assistants pick pages that make citation easy.
- Question in the h1 or h2. Example: “Can I import CSV orders into YourProduct?” on /guides/import-csv-orders.
- Answer in the first two sentences. State yes or no. Name the product the same way every time. Example: “Yes, YourProduct imports CSV orders. Upload a UTF‑8 file with headers.”
- A list or a table with the facts. Steps, limits, supported formats, versions, or prices if relevant.
- Consistent names for your product, plans and features. Pick one form. Use it everywhere.
- A visible date and an author. Example under the title: “Updated 14 September 2026 · by Alex Tan.”
- Sources at the end. Link the standards, docs and any third‑party numbers you quote.
- A short URL. Example: yourproduct.com/guides/import-csv-orders.
- Titles and headings under about 60 characters where possible, so they scan well.
A fixed page looks like this when done: the h1 is the question, the first paragraph answers in two sentences, then a steps list, then a small table of file limits, then sources, date and author shown near the top.
A worked example: change a vague guide into a cited one
Take a common founder page: “Data import guide” at /guides/data-import. It opens with three paragraphs of product story and screenshots. The method is buried. Assistants skip it.
Pick one buyer question and make it the h1
Rename the page: “How to import orders from CSV into YourProduct”. Put it at /guides/import-csv-orders.
Write the two‑sentence answer first
“Yes, YourProduct imports CSV orders. Upload a UTF‑8 CSV with headers order_id, customer_email, total, created_at.”
List the steps
1. Go to Settings, Imports. 2. Choose Orders. 3. Upload the file. 4. Map headers. 5. Validate. 6. Start import. 7. Review the log.
Add a small facts table
Include limits, accepted headers, date formats, time to process. Keep it scannable.
Name everything consistently
Write “YourProduct” everywhere, not “YP” or “our platform”. Use the same plan names used on /pricing.
Add date, author, and sources
Place “Updated 14 September 2026 · by Alex Tan” below the title. Link to your CSV schema and the RFCs you cite.
| Fact | Value | Where shown |
|---|---|---|
| Maximum file size | 50 MB | Facts table |
| Accepted MIME types | text/csv | Facts table |
| Encoding | UTF‑8 only | Facts table |
| Date format | YYYY‑MM‑DD | Facts table |
| Headers required | order_id, customer_email, total, created_at | First answer + facts table |
| Processing time | About 2 minutes for 10k rows | Facts table |
Before: a generic guide with a soft intro and no facts block. After: a question in the h1, a two‑sentence answer, seven steps, and a facts table the assistants can quote line by line.
What the assistants have in common
Generative engine optimisation, GEO, aims to be named and cited in answers written by AI systems. The term was popularised by a 2023 paper by Aggarwal and colleagues. The playbook is now clear enough to use.
- They retrieve from a web index. It may be Bing’s, Google’s, or their own crawl.
- They cite pages that answer the question early and plainly, with headings that match the query.
- They prefer lists and tables they can quote without rewriting.
- They reward consistent naming of products and people.
- They show a visible date and author if you include them.
- They name products they find on lists, roundups and comparisons they already trust.
- They cannot read content rendered only after JavaScript runs.
- They cannot be paid for a citation.
What each assistant reads and how to be visible
You do not submit pages. You make sure the right crawlers can read them, then you test by asking the assistants. Here is what is documented as of 2026.
| Assistant | Index and crawling | How it cites | Notes you can act on |
|---|---|---|---|
| ChatGPT search | Mix of search partners, Bing among them, and OpenAI’s OAI‑SearchBot | Inline citations and a Sources panel with links | Launched Oct 2024 to paid users, then to all. The crawler does not execute JavaScript. Allow OAI‑SearchBot to appear, you can still disallow GPTBot if you wish to avoid training. |
| Perplexity | PerplexityBot index, Perplexity‑User fetches on demand | Numbered citations and sources list | Prefers pages that answer in the first paragraphs with clear headings and recent dates. Has Pro Search and a publisher programme. |
| Google AI Mode and AI Overviews | Google Search index, query fan‑out and Gemini models for AI Mode | Links inside the AI answer and organic results below | Cites pages ranking in Google already. No separate crawler or opt‑out beyond snippet controls. Clicks arrive as google.com in analytics. |
| Gemini (app) | Grounds with Google Search when it searches | Shows links when grounded | Google‑Extended controls training and grounding in Gemini apps, not Google Search or AI Overviews. |
| Claude | Reportedly built on Brave Search index, Claude‑SearchBot indexes, Claude‑User fetches | Inline citations with links | Web search arrived March 2025. Allow Claude‑SearchBot to be in the index. |
| Microsoft Copilot | Bing index, Bingbot crawler | Citations from Bing’s results | Use Bing Webmaster Tools and IndexNow to ensure coverage. Bingbot is the crawler. |
Other crawlers appear in your logs and robots file. Applebot and Applebot‑Extended, CCBot, Bytespider, meta‑externalagent and meta‑externalfetcher, Amazonbot, cohere‑ai, DuckAssistBot. These are part of training and other services. Robots.txt is honoured voluntarily by vendors. Blocking a training crawler does not remove pages already collected.
Let the right crawlers in
If answering crawlers cannot fetch your pages, assistants will not cite you. Allow the search and answering bots in robots.txt. You may still choose to block training bots.
- Allow answering and search crawlers: OAI‑SearchBot, Claude‑SearchBot, PerplexityBot, Bingbot, Googlebot.
- Decide on training crawlers: GPTBot, ClaudeBot, Google‑Extended, Applebot‑Extended, CCBot, meta‑externalagent, and others.
- Keep sitemaps open. Ensure your main pages are linked and indexable.
- Test access with your server logs or a robots tester.
# Example robots.txt allowing answering/search, limiting training
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
User-agent: GPTBot
Disallow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /
User-agent: ClaudeBot
Disallow: /
User-agent: PerplexityBot
Allow: /
User-agent: Perplexity-User
Allow: /
User-agent: Googlebot
Allow: /
# Control Gemini training and grounding in Gemini apps
User-agent: Google-Extended
Disallow: /
User-agent: Bingbot
Allow: /
User-agent: Applebot
Allow: /
# Apple Intelligence training control
User-agent: Applebot-Extended
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: meta-externalagent
Disallow: /
User-agent: meta-externalfetcher
Allow: /
User-agent: *
Disallow:Be on the lists assistants already cite
Assistants name products they see on trusted lists and comparison pages. There is no submission form. You earn mentions where buyers already look.
Find the cited lists for your topic
Ask each assistant the buyer questions your page should answer. Note the lists, roundups and comparison pages they link.
Qualify for inclusion
Match the criteria visible on those pages. For example, a clear pricing page, a free trial, screenshots, a public changelog, or an API page.
Pitch with facts, not prose
Email the editor with your product’s name, one‑line description, your primary page URLs, and the facts they list in their table.
Make your own comparison page
If the gap is obvious, build /comparison/yourproduct-vs-rival and /alternatives/yourproduct-alternatives with the same citation‑ready pattern.
A fixed page here is a comparison with the rival named in the h1, a two‑sentence summary, and a table of differences that quotes cleanly. Link both ways from /pricing and /guides.
What to avoid and why it costs citations
- Answers hidden in JavaScript. OAI‑SearchBot does not execute scripts. Content that renders after load will not be read.
- Marketing prose before the answer. Assistants skip long intros. Put the answer first.
- Unnamed authors. Assistants prefer pages with a real byline and a visible date.
- Shifting names. Do not alternate “YourProduct CRM”, “Your Product”, and “our platform”. Pick one.
- Walls of images and PDFs. If the text is vital, write it in HTML.
- Out‑of‑date pages. Out‑of‑date facts lose trust. Update and show the date.
A fixed page removes the soft intro, shows the answer first, puts the facts in a list, and gives a person’s name and an update date at the top.
Test by asking the assistants
Measuring GEO means asking the assistants the questions a buyer asks and reading who is named. No assistant publishes a report of where you were cited, as of 2026.
Write the question list
Use your top help topics, feature gaps, and competitor claims. Example: “How to migrate from RivalApp to YourProduct” and “YourProduct CSV import headers”.
Ask on a schedule
Monthly at minimum. Ask ChatGPT search, Perplexity, Claude, Gemini, and Microsoft Copilot. Log the names and links each one cites.
Track analytics referrers
Group chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Google AI Overviews and AI Mode appear as google.com.
Use tracking tools if you need scale
Tools like Profound, Peec AI, Otterly, Scrunch, and the AI toolkits of Semrush and Ahrefs, and DataForSEO’s LLM mentions, automate the asking.
Fix, re‑ask, and note movement
When you change a page, re‑ask in a week. Keep a simple sheet of which question now cites you and where you rank in the sources list.
A fixed habit is a one‑page log with your ten questions and a column for each assistant. Colour the cells when you are cited. This keeps you honest without noise.
Page types to prioritise on a product site
Start with the pages most likely to be retrieved for buyer questions. Each should follow the same citation‑ready pattern.
- Feature how‑to guides. Example: /guides/import-csv-orders and /guides/api-keys.
- Comparison and alternatives. Example: /comparison/yourproduct-vs-rival and /alternatives/yourproduct-alternatives.
- Migration pages. Example: /migrate/from-rivalapp to answer “How to migrate from RivalApp to YourProduct”.
- Pricing and limits. Example: /pricing with a small table of plan limits assistants can quote.
- Security and compliance. Example: /security with a table of certs and data locations.
- Changelog and release notes. Example: /changelog with dates and one‑line summaries assistants can ground to.
A fixed /pricing page has a short answer to “What does YourProduct cost?”, then a table. Use the same plan names on every page to avoid confusion in citations.
Formatting that gets quoted
Write for quotation. Keep facts atomic. Place them where an assistant can lift a line without harm. Avoid ambiguity and synonyms for key entities.
- Use ordered steps for procedures. Keep each step one action.
- Use short tables for limits and comparisons. One fact per cell.
- Repeat the exact product and plan names. Do not alternate them.
- Put version numbers, dates and units near the fact they qualify.
- Prefer simple tenses and short sentences. Avoid nested clauses.
- Link authoritative sources you cite. Assistants can show them with yours.
A fixed limits section lists the API rate limit on its own line. This is easy to quote, unlike a paragraph that buries the number mid‑sentence.
How to fix JavaScript‑heavy pages
Several assistants do not execute JavaScript, and some search crawlers execute less than a browser. If your answer only renders client‑side, it will be missed.
Render the core answer in HTML on first load
Server‑side render or pre‑render the first two sentences, the steps list, and the facts table.
Avoid gating the content behind interactions
Do not hide the answer behind tabs or accordions that require scripts to populate.
Check what OAI‑SearchBot would fetch
Use View Source or a curl request. If the text is not in the raw HTML, crawlers may not see it.
Provide clean URLs for each question
Avoid hash‑routed single pages. Give each guide its own path like /guides/import-csv-orders.
A fixed SPA page exposes the h1, the two‑sentence answer, and the facts list in the initial HTML, even if the rest hydrates later.
Questions
Allow OAI‑SearchBot in robots.txt so your pages can be in ChatGPT’s search index. Write pages that answer the question in the first two sentences with clear headings, lists and a visible date and author. Then ask ChatGPT search your buyer questions and see if it names you. If not, refine the page and try again.
Answer engine optimisation, AEO, came from the featured‑snippet and voice‑assistant years. It aimed to be the answer rather than a blue link. Generative engine optimisation, GEO, is the same aim applied to AI systems like ChatGPT, Perplexity, Google AI Overviews and AI Mode, Claude, Gemini and Copilot. The overlap is now large.
No. There is no paid placement for citations. Assistants retrieve from web indexes and cite pages that answer directly. Being on trusted lists and comparisons that they already cite is the most reliable route.
No. Training crawlers and answering crawlers are separate. Allow the answering and search crawlers such as OAI‑SearchBot, Claude‑SearchBot, PerplexityBot, Googlebot and Bingbot. You may still choose to disallow GPTBot, ClaudeBot, Google‑Extended or Applebot‑Extended if that suits your policy.
Keep a list of buyer questions and ask each assistant on a schedule. Log who is cited and the URLs. Track referrers in analytics: chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. If you need scale, use tools that automate the checks, then fix pages and re‑ask.
If the content only appears after JavaScript runs, it may be missed. Some crawlers do not execute scripts. Put the answer, steps and facts in the initial HTML. Avoid hiding vital text in images or PDFs if you want it cited.
Check my site, free
Want to see if your pages fit this pattern and which rivals assistants cite instead, from just your URL in about thirty seconds, free, showing three findings whole? Try Porteur’s check.
- Free check, no card
- Read-only, your own accounts
- Readable by your agent
Read next
- GuideGenerative engine optimization (GEO): how to be named by AI answers
- GuideHow ChatGPT search picks and cites sources
- GuideHow Perplexity picks its sources, and how to become one
- GuideGoogle AI Mode: what it is and how a small site gets cited
- Guiderobots.txt for AI crawlers: GPTBot, ClaudeBot, PerplexityBot and what to allow
- GuideHow to track your AI visibility without a subscription
- GlossaryChatGPT search
- GlossaryGenerative engine optimization (GEO)