How to get your website cited by ChatGPT, Claude and Google's AI answers

To improve your eligibility for AI search citations, keep useful public pages accessible and follow each provider’s current search controls. Google requires normal Search eligibility; OpenAI and Anthropic document separate search and training crawlers. Eligibility is not a promise that a model will select your page. Start with accurate, original answers to a real reader problem, then check access, indexing and measurement before changing copy.

The question your best page already answers

Somewhere right now, a person is asking an AI assistant the exact question your best article answers. The assistant writes a tidy paragraph, adds a few links, and moves on. Your site is in those links, or it is not.

A made-up example of how sites drop out without noticing. A travel agency in Dubai installs a security plugin with a "block AI bots" switch, because it sounds sensible. Months later the owner asks ChatGPT for the best desert safari operators and sees competitors cited, never her own detailed guides. Her robots.txt is fine; the plugin's firewall was quietly turning away every AI crawler, including the ones that power search.

Nobody can promise you a citation, including us. What you can do is remove the blockers, make the right choices about training, and give these systems a page worth quoting. Most of it you can check yourself in fifteen minutes.

What Google says

Google’s current AI optimization guide applies ordinary Search quality principles to AI Overviews and AI Mode. It recommends useful, distinctive content rather than a special writing formula. No special AI schema or llms.txt file is required for these Google features. Clear explanations, relevant visuals and evidence help readers; none forces a citation.

For Google’s AI search features, a supporting page must be indexed and eligible for a Search snippet. Keep important content accessible as text, use crawlable internal links, and ensure structured data matches the visible page. Meeting these conditions establishes eligibility, not guaranteed inclusion or a higher rank.

Check page-level controls such as noindex, nosnippet, max-snippet and data-nosnippet. They affect what Google may index or use in previews; data-nosnippet applies only to marked content. Separately, in Search Console Settings, inspect the effective Search generative AI inclusion control and any inherited parent setting. Google rolled this control out worldwide by 31 August 2026. Review the owner’s intended policy before changing it: ordinary Search inclusion, AI-feature inclusion and model training are different decisions.

Crawlers: one switch per job

Each AI company runs more than one crawler, and each has its own job. Think of them as different delivery drivers for the same company: one collects material for the warehouse (training), one fetches a parcel when a customer asks (a user request), and one builds the catalogue (search). You can turn away one without turning away the others.

OpenAI

• OAI-SearchBot is used "to surface websites in search results in ChatGPT's search features." OpenAI recommends allowing it in robots.txt, and allowing requests from its published IP ranges, to help ensure your site appears.
• GPTBot crawls content that may be used to train OpenAI's foundation models. Blocking it is your training opt-out.
• ChatGPT-User visits a page when someone asks ChatGPT something. OpenAI notes that because these actions are initiated by a user, robots.txt rules may not apply.

OpenAI says it can take around 24 hours from a robots.txt update for its systems to adjust.

Anthropic

• ClaudeBot collects web content that could contribute to training. Disabling it signals that your future content should be excluded from training.
• Claude-User fetches pages when people ask Claude questions. Blocking it can reduce your visibility for user-directed web search.
• Claude-SearchBot improves search result quality. Anthropic says blocking it may reduce your visibility in search results.

Anthropic notes its bots honour robots.txt, and that blocking its IP addresses can backfire because it stops the bots reading your robots.txt at all.

Google

Googlebot supports Google Search. Google-Extended is a separate robots control token for the Gemini training and grounding uses described in Google’s current documentation; it is not a separate HTTP user-agent name to match in logs. Google states that it is not a Search ranking signal. For AI Overviews and AI Mode participation, inspect Search Console’s dedicated Search generative AI control rather than assuming Google-Extended is that switch.

Research observations about AI retrieval do not by themselves prove that a crawler preference caused a ranking or citation change. Choose training and grounding preferences deliberately, using the provider’s documented controls. Do not weaken those preferences merely to chase an unproven SEO benefit. If visibility changes, investigate indexing, access, effective AI inclusion settings, content and measurement separately.

Robots.txt recipes you can copy

Robots.txt lives at yourdomain.com/robots.txt. Google's rules choose the most specific matching user-agent group; multiple equally specific groups are merged. A specific group is not combined with the general User-agent: * group. Keep the relevant path restrictions in each applicable specific group.

Recipe A: welcome search, opt out of training

This simplified example welcomes search crawlers while retaining the WordPress admin-path restrictions in their specific groups. It opts out of GPTBot and ClaudeBot. Adapt it to your site’s existing rules instead of replacing the whole file; robots.txt is a crawl preference, not an access-control barrier. Replace the sitemap address with yours:

User-agent: OAI-SearchBot
User-agent: Claude-SearchBot
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.php

User-agent: GPTBot
User-agent: ClaudeBot
Disallow: /

User-agent: *
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.php

Sitemap: https://example.com/sitemap.xml

Add a Google-Extended group with Disallow: / only if that matches your choice about its documented Gemini training and grounding uses. This is distinct from the Search generative AI control. Keep Googlebot access aligned with your ordinary Search goals and retain all unrelated site restrictions.

Recipe B: welcome everything

If you are happy for your content to be used for search and training, you only need the general group and your sitemap. Search crawlers then follow the * rules.

User-agent: *
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.php

Sitemap: https://example.com/sitemap.xml

What to avoid

A line Disallow: / under User-agent: * blocks the whole site for every crawler that has no group of its own, including Googlebot. It is a common leftover from a staging site that went live. In WordPress, the Discourage search engines from indexing this site box under Settings > Reading is worth checking too.

The fifteen-minute check

1. Read your robots.txt

Open yourdomain.com/robots.txt in a browser and look for any group that blocks OAI-SearchBot, Claude-SearchBot or Googlebot, or a site-wide Disallow: /.

2. Check that Google can index you

In Search Console, inspect an important canonical page to see Google’s indexed status and check robots.txt fetch errors. The indexed inspection is not a live crawl or a promise of ranking. Also inspect the effective Search generative AI setting, including inheritance from a parent property, before diagnosing missing AI visibility.

3. Check your firewall and security plugin

Look for settings such as "block AI bots" or aggressive bot-fighting modes in your security plugin or CDN. A robots.txt that says yes does not help if a firewall says no. OpenAI publishes the IP ranges for OAI-SearchBot at https://openai.com/searchbot.json so you can allow them.

4. Test what a bot sees

This request pretends to be OAI-SearchBot by its user agent name:

curl -s -o /dev/null -w "%{http_code}\n" -A "OAI-SearchBot" https://example.com/

A 403 means this particular request was denied; it does not establish that the user-agent name caused the denial. The source IP, WAF rule, rate limit or challenge may matter. Compare with a normal request and verified real-crawler logs. A 200 from a spoofed user agent does not prove the real crawler can fetch the page or will cite it.

5. Look for real visits in your logs

Download your access log (on cPanel, from the Raw Access page) and search it:

grep -E "OAI-SearchBot|Claude-SearchBot|Googlebot" access.log | tail -n 20

A matching user-agent name is only a clue because it can be spoofed. Verify the source using the provider’s published identification method before calling it a genuine crawler. Then compare response status and requested path. A successful fetch confirms access for that request, not indexing or citation.

What to write so a machine can quote you

A research paper, "GEO: Generative Engine Optimization" (Aggarwal and others, accepted at KDD 2024), tested ways of editing pages so generative engines would feature them more. Its abstract reports that the methods can boost visibility by up to 40%. Adding quotations, adding statistics and citing sources were among the most effective changes, while keyword stuffing, the old search trick, did not help.

In the paper's multi-website experiment, where all source pages were optimized together, Cite Sources changed the reported visibility metric by +115.1% for fifth-ranked sources and -30.3% for first-ranked sources. These are experimental relative changes by rank, not percentage-point gains, predicted traffic growth or a guarantee for an individual website.

Keep the caveat in view: these were tests on a research benchmark, not promises about how ChatGPT, Claude or Google behave today. Treat them as a reason to write well, not as a trick. In practice, that means a few habits:

• Answer first. When the reader needs a direct answer, put it near the start and then explain it. This is a readability choice, not a required sentence count or citation formula.
• Source your numbers. Name the source and link to it whenever you state a figure.
• Quote real people and documents with attribution. Never invent a quote.
• Use descriptive headings and add FAQs only when they resolve real follow-up questions. Do not force question headings or a FAQ quota into every article. Google ended FAQ rich results in May 2026; useful visible Q&A can still help a reader, but it is not a promise of extra Search appearance.
• Show who stands behind the page: an author, a reviewed or updated date, and an about page that explains your expertise.
• Keep it in text. Important information locked in images or PDFs is harder for any crawler to use.
• Use accurate structured data, such as Article markup with the author and dates, where it matches the visible page.

Pages worth building first

Some kinds of page are naturally easier to quote, because they answer a specific question completely. If you are deciding what to write next, start here:

• Clear definitions and explainers for terms your customers search for, with a one-sentence answer at the top.
• Step-by-step guides with numbered steps, exact settings and what to do when something goes wrong.
• Comparisons with stated criteria, such as price, features and who each option suits, rather than vague "top ten" lists.
• Local and service pages with specifics: the areas you serve, opening hours, typical prices or price ranges, and how to contact you.
• Original information nobody else has: your own survey results, case studies with real (permitted) numbers, or a checklist from years of doing the work.

How to tell whether it is working

You cannot see inside these systems, but you can watch for signs.

• Referral traffic: inspect visits attributed to domains such as chatgpt.com, claude.ai, gemini.google.com or perplexity.ai. These show referred visits where attribution is available, not every citation or impression. A copied link, missing referrer or in-app behavior can obscure attribution. Evaluate useful reader actions and qualified enquiries, not just raw referral counts.
• Search Console: the dedicated generative AI performance report shows impressions from AI Overviews and AI Mode, with page, country, device and date breakdowns. It may be absent when there are insufficient impressions. These data overlap overall Web reporting, so do not add the totals together or label them isolated AI clicks or conversions. Keep comparable complete reporting periods and record when each page changed.
• A repeatable spot check: keep a small set of genuine customer questions and record the date, product, locale and visible cited URLs when you test them. Results vary across runs and users, so a single answer is not a stable ranking measurement. Use this qualitative evidence alongside Search Console, referral data and reader outcomes.
• Your logs: the log search above shows whether search crawlers keep visiting after you make changes.

Make the site technically easy to read

• Speed and availability matter. A page that times out for a crawler is a page that cannot be quoted. Our LiteSpeed Cache guide covers speed for WordPress sites.
• Use HTTPS everywhere and a single canonical address, as our HTTPS redirect guide explains.
• Keep an XML sitemap and list it in robots.txt.
• Link related pages together, so crawlers find your deeper guides from your popular ones.

Check your site's current robots.txt and sitemap rather than assuming a previous configuration is still in place. Protect private CMS routes with access control, not just crawler directives. If you are weighing hosting for a site like this, our piece on what AI-powered hosting actually changes explains where automation helps and where people still matter.

The one message worth forwarding

Copy this into a message to your web developer or host:

Ask your developer or host to verify search-crawler access without relaxing unrelated protections, check unintended page-level snippet/indexing restrictions, and confirm the effective Search generative AI setting. Decide separately whether to opt out of each provider’s training and grounding uses. Ask for evidence from the actual configuration and verified logs, not just a successful spoofed-user-agent request.

It takes them a few minutes and clears problems that are easy to miss from the outside. Then spend the rest of your effort on the part nobody can automate: writing a page that is worth citing.