AI Search and Your Website: Practical Steps That Still Matter
ChatGPT, Perplexity, Claude and Google's AI answers each find pages in their own way. Five steps that help with all of them, and the fads to skip.
Google’s documentation on AI Overviews and AI Mode has one line worth printing out: “There are no additional requirements to appear in AI Overviews or AI Mode.”
That hasn’t stopped a whole industry of “GEO”, “AEO” and “LLM SEO” from appearing. Some of it helps. Much of it is old SEO with a new name.
Clicks are getting harder, though. In a US study published in July 2025, Pew Research found Google users clicked a regular result on 8% of visits with an AI summary, against 15% without.
The short version: AI search isn’t one system. Let the search bots in, make pages easy to index, and publish facts nobody else has.
Then keep your name consistent across the web, and measure citations and referrals instead of chasing a rank.
The 5 steps at a glance
| # | Step | About |
|---|---|---|
| 1 | Decide which AI crawlers to let in | 30 min |
| 2 | Make your pages indexable and easy to understand | 1 hour |
| 3 | Publish facts nobody else has | Ongoing |
| 4 | Be consistent about who you are, and earn mentions | 1 hour, then ongoing |
| 5 | Measure AI citations and referrals | 30 min a month |
1. Decide which AI crawlers to let in
Most AI companies now run separate bots for search, training and user requests, and you can treat each one differently. Here’s how their own documentation describes them:
| Company | Search (shows and links your site) | Training | Fetches a page when a user asks |
|---|---|---|---|
| OpenAI | OAI-SearchBot | GPTBot | ChatGPT-User |
| Anthropic | Claude-SearchBot | ClaudeBot | Claude-User |
| Perplexity | PerplexityBot | None; Perplexity says PerplexityBot isn’t used for training | Perplexity-User |
| Googlebot, since AI Overviews and AI Mode come from Google Search | Google-Extended, a robots.txt token for Gemini | Not applicable |
The search bots decide visibility. OpenAI says sites that block OAI-SearchBot “will not be shown in ChatGPT search answers”.
Anthropic gives a similar warning about Claude-SearchBot.
Training is a separate choice. You can block GPTBot or ClaudeBot and stay in their search, because, as OpenAI puts it, each setting is independent of the others.
Google-Extended doesn’t affect inclusion in Google Search, Google says.
Google also has a Search generative AI control in Search Console for AI Overviews and AI Mode.
The rules keep moving. Microsoft has split AI training from AI grounding in Bing’s controls, and Google plans to support one shared robots.txt rule for AI training.
Check your CDN as well. Firewall and bot settings can block AI crawlers even when robots.txt allows them.
Do you run ads? Search bots can bring citations and visits, with no promise of either, while blocking them keeps your pages out of those answers.
Decide per company, by what each one sends you.
Check it: read your robots.txt for each bot above, check your CDN’s bot settings, and look for the search bots you’ve allowed in your server logs.
2. Make your pages indexable and easy to understand
AI answers draw on search indexes. Google’s AI features use pages from Google Search.
For ChatGPT, Ahrefs studied 1.4 million prompts and found 88% of cited URLs came from its search retrieval.
So the basics carry over. Your pages should be indexed, with the main content as real text, a clear canonical URL and links from the rest of the site.
The technical SEO audit checklist checks all of those.
Write for someone who wants the answer. Put it where they’ll see it.
Use headings that match real questions, and tables for comparisons. Google also says “There’s no requirement to break your content into tiny pieces for AI.”
Check it: run five key pages through Google’s URL Inspection and Bing Webmaster Tools. Each should be indexed, canonical to itself, and show its main text in the rendered page.
3. Publish facts nobody else has
An AI answer is assembled from sources. The page with the original number is the one worth citing.
That means first-hand data, prices you checked, tests you ran, dates things changed, and an author who knows the subject.
Keep it current, too. An old price quoted in an AI answer reflects on you.
Seen the “40%” figure? It’s from a 2024 paper on generative engine optimization, which found citations, quotes and statistics could lift visibility “by up to 40%” in its own benchmark.
Treat it as a lab result.
Check it: for each key page, name one fact readers can’t get from your competitors, with its source and the date you checked it.
4. Be consistent about who you are, and earn mentions
AI tools build a picture of your business from many sites. If your website, Google Business Profile and LinkedIn give different names or descriptions, the picture blurs.
Make them match. Organization structured data on your homepage helps, as long as it agrees with what’s visible.
Then earn mentions where AI tools look: editorial coverage, reviews, real answers in communities your customers use.
Our link hub lists the ones we’ve checked.
Which sources get cited varies by tool, topic and month. Chasing one platform is a weak plan, and a paid spot on a “best of” list is just a line on someone else’s page.
Check it: compare your business name, contact details and one-line description across your site and your five most-visited profiles, and fix every mismatch.
5. Measure AI citations and referrals
You can measure more than most guides admit:
- Search Console has a generative AI performance report with your impressions in AI Overviews and AI Mode, by page, device and country.
- Bing Webmaster Tools has an AI Performance report showing which pages Microsoft Copilot and partner experiences cite.
- Google Analytics shows referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, though tagging is patchy: Gemini seems to have dropped UTM tags.
Some AI-driven visits arrive with no referrer at all. People copy a link or search your name later, so part of it lands in “direct”.
A monthly test helps. Ask the same ten customer questions in each AI tool and note whether you were named.
Treat it as a sample. Answers change from one day to the next.
Check it: a saved analytics report of AI referrals by source, and a dated monthly log of your ten test questions.
What to ignore
- llms.txt for Google, which says these files “neither harm nor help” visibility in Google Search (other tools may read them)
- special “AI schema”
- “rank #1 in ChatGPT” promises, since AI answers have no fixed public ranking
- rewriting every page into tiny chunks
Why it matters
AI search changes where clicks come from. It doesn’t change what earns them: pages worth citing, indexed, clear, original and consistent.
So skip the new acronyms. Check your bot settings once, keep doing the SEO that works, and spend your effort on facts people will want to quote.
The tools to see AI impressions and referrals exist now. Judge every AI tactic by what it sends you.
Common Questions
Is generative engine optimization different from SEO?
Mostly, no. Google says SEO best practices remain relevant for AI Overviews and AI Mode, with no additional requirements. What’s new is choosing which AI crawlers to allow and measuring AI citations and referrals separately from regular search traffic.
Does llms.txt help with AI search?
It doesn’t help in Google. Google says llms.txt files neither harm nor help visibility in Google Search, which ignores them. Some other AI tools may read the file, so keeping one is harmless, but it isn’t a ranking lever.
Should I block AI crawlers from my website?
Decide bot by bot. Training bots like GPTBot and ClaudeBot can be blocked without leaving AI search, while blocking search bots like OAI-SearchBot or Claude-SearchBot removes your pages from those answers, according to OpenAI and Anthropic.
