AI SEO Guide: A Step-by-Step Tutorial for Your First Pass

An AI answer about your brand comes from one of two places. Either the model answers from what it learned in training, or it searches the web first and builds the answer from pages it retrieved. It helps to split this into four stages.

The engine also decides per question whether to search at all. ChatGPT searches at the model's discretion, Claude.ai search is a user toggle, and Google AI Overviews often don't trigger. So the same question can get a searched answer one day and a memory-only answer the next.

What this means for you: the access stage is the one you fully own. A July 2026 critical survey of 45 GEO studies (arXiv 2607.14035) found that no reviewed technique shows a stable, cross-platform causal effect on discoverability. The work that holds up is making sure nothing stops engines from reaching and reading you.

Be clear about the split before you spend money.

In your control

• Whether each AI crawler is allowed in robots.txt.
• Whether your firewall or CDN lets those crawlers through.
• Whether your key content is in the HTML before JavaScript runs.
• Whether your pages are indexable and allowed a snippet in Google and Bing.
• Whether your own site states your facts (prices, product names, locations) clearly and consistently.

Not in your control

• What a model learned before its training cutoff.
• Whether an engine decides to search for a given question.
• Which queries the engine actually runs. OpenAI rewrites prompts into its own searches, and Google can fan out across subtopics.
• Which retrieved page gets cited, and in what order.

Nobody can promise you a citation, a ranking or a score. What you can do is remove every block on your side, then watch whether the answers change. The steps below follow that order.

Before you fix anything, write down the baseline. Pick three to five questions a buyer would actually ask, such as "best accounting software for freelancers" or "what does [your brand] cost". Ask them in ChatGPT, Gemini, Perplexity, Claude and Google, logged out where you can, since memory and custom instructions change results.

For each answer, record three things:

1. Whether your brand is named at all.
2. Whether any stated fact is wrong: an old price, a retired product, the wrong city.
3. Which sites the engine cited instead of yours.

Wrong facts matter most. They're specific, you can trace them to a source, and they're what buyers act on. Cited third-party sites matter too. When engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer, though nobody has measured a guaranteed effect.

If you want a quick snapshot, the free checker on this site runs up to three buyer questions and scores whether you're named. It's a one-time check, not ongoing tracking.

Next move: save the answers with the date. You'll compare against them in Step 5.

Access has two gates, and most sites only check one.

Gate one: robots.txt, bot by bot

Each AI company uses separate crawlers for separate jobs. OpenAI's crawler documentation says GPTBot collects content for training, OAI-SearchBot is used for ChatGPT search, and "each setting is independent of the others". Blocking GPTBot doesn't remove you from ChatGPT search. Allowing it doesn't get you into search either. The common mistake is a blanket AI block that silently removes search eligibility. For example, this lets ChatGPT search in while opting out of training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Also check PerplexityBot, Claude-SearchBot and Meta-WebIndexer. There's no robots.txt token for Google AI Overviews or AI Mode; Googlebot rules cover all Google Search surfaces. Our per-bot robots.txt guide lists each token and what blocking it does.

Gate two: the firewall and CDN

A firewall or bot-management tool can return a 403 to a crawler that robots.txt allows. OpenAI's help centre says eligibility needs both: allow OAI-SearchBot and confirm your host or CDN allows OpenAI's search crawler traffic. Cloudflare also changed its defaults on 2026-09-15, blocking Training and Agent crawlers on ad-displaying pages for new domains and existing free-tier customers. You can lose access without editing anything.

Next move: request a page as each crawler and compare the response with your robots.txt. The AI crawler access audit playbook walks through it, and the AI Crawler Checker shows which crawlers your firewall turns away.

This is the strongest finding in the evidence. Vercel's crawler study reported that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them. They can't read client-side rendered content." Independent testing as of June 2026 found the same for GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot and others. Googlebot and AppleBot are the exceptions, because they render pages.

So if your pricing table, product list or service areas load through client-side JavaScript, Google may see them while ChatGPT and Perplexity see an empty shell.

How to check:

1. Open a key page with JavaScript disabled, or view the raw page source.
2. Compare it with the page as a visitor sees it.
3. List every fact that only appears in the second version.

Fix it with server-side rendering or prerendering for the content that matters. Accordions, tabs and "load more" buttons that fetch content later have the same problem.

While you're there, state specific facts plainly. A price, a number or a one-line definition on a fetchable page gives the engine a sentence it can quote. It's the only on-page content change with a positive experimental signal.

Next move: fix your pricing, product and about pages first, since those hold the facts engines get wrong most often.

An engine can't cite what its index doesn't hold. Google's AI features documentation says a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link in AI Overviews or AI Mode, with "no additional technical requirements".

Check these in Google Search Console:

• Your key pages are indexed.
• No template adds noindex, nosnippet or max-snippet:0 by accident.
• No data-nosnippet attribute hides content you want shown. Google and Bing both treat it as a suppression control.
• The generative AI inclusion control, where you have it. Inclusion is on by default, but child properties inherit the parent's setting.

Then check Bing Webmaster Tools. Bing's index backs Microsoft Copilot. Otterly's 2026 experiment with two AI-generated sites shows why both matter: Google deindexed the sites after about 14 and 90 days, Bing kept them, and Copilot produced 2,718 of the 3,934 citations Otterly tracked for one site. Google's AI surfaces almost never cited them. One healthy index tells you nothing about the other. Our Copilot visibility checklist covers the Bing side.

Next move: fix any accidental suppression, then request reindexing of the corrected pages.

Go back to your Step 1 questions after your fixes have had time to be recrawled, and compare engine by engine. A single blended score hides the pattern you need. In Otterly's test, one engine cited the sites thousands of times while four barely did.

Keep your expectations matched to what can be measured as of October 2026:

• Google's generative AI report in Search Console shows impressions only. It has no clicks, CTR or position data.
• No engine exposes a rank inside an AI answer, so any "AI position" number has nothing to be checked against.
• Bing Webmaster Tools' AI Performance report shows grounding queries, the searches the AI actually ran. It's the only first-party view of those.
• Any score averages answers given with and without live search, because the engine decides when to search.

Track presence, wrong facts and cited sources per engine, and note the date. Our guide to measuring AI share of voice sets up a repeatable method, and the ChatGPT visibility checklist goes deeper on one engine.

Next move: repeat the same questions monthly, and treat a new wrong fact as a fix target.

Much AI SEO advice still repeats tactics that primary sources contradict. Skip these:

• Adding schema to get cited. Google says structured data isn't required for generative AI search. Ahrefs' matched-control study of 1,885 pages found AI Overviews citations fell 4.6% after adding schema. Existing markup does no harm, but it isn't a lever for AI answers. FAQ rich results were also deprecated in May 2026.
• Publishing an llms.txt file. Google says Search ignores it, and no other engine documents reading it.
• Chasing an E-E-A-T score or domain authority. Google says E-E-A-T isn't a specific ranking factor, and domain authority is a third-party metric no engine computes.
• Rewriting for an "AI-friendly" style or chunking pages. Google rejects both and says there's no ideal page length. Our digest of Google's generative AI optimization guide covers its full list.
• Mass-publishing AI content. Google's spam policy targets many pages made to manipulate rankings without value, and removal comes with no warning.
• Hidden text or prompt injection aimed at AI crawlers. It's illegitimate, and it risks penalties and reputational damage.

For what to do after this first pass, see GEO optimization: a practical order of work.

Is AI SEO different from regular SEO? Partly. Google's AI Overviews and AI Mode use the same index and eligibility rules as Google Search, so normal indexing hygiene carries over. The differences are elsewhere. Non-Google AI crawlers mostly don't run JavaScript, each AI company uses separate crawlers for training and search, and engines like Copilot draw on Bing's index. AI SEO adds those checks on top of standard SEO.

How long does it take for fixes to show up in AI answers? Nobody publishes a timeline. Searched answers can change once crawlers refetch and reindex your corrected pages. Answers given from model memory won't change until a new model is trained, which can take many months. ChatGPT's knowledge cutoffs, for example, differ by model. Recheck your baseline questions monthly and compare engine by engine.

Should I block GPTBot to protect my content? That's a business choice, and it's separate from search visibility. OpenAI's documentation says blocking GPTBot signals your content shouldn't be used for training, while OAI-SearchBot controls ChatGPT search. You can block one and allow the other. Nothing published measures whether blocking training affects how models describe your brand later.

Can a small team do this without an agency? Yes, for a first pass. Checking robots.txt, testing crawler responses, comparing raw HTML with the rendered page, and reviewing Search Console and Bing Webmaster Tools are all mechanical tasks. The harder part is ongoing measurement across engines. Start with the access and rendering checks, since those have the strongest evidence and need no ongoing spend.