What AI SEO means, and the three things it can refer to

People say AI SEO and mean one of three things.

• Using AI to do SEO work: research, briefs, drafting, audits, QA.
• Generative engine optimisation, GEO: being named and linked in AI answers from ChatGPT, Perplexity, Google AI Overviews and AI Mode, Claude, Gemini and Microsoft Copilot.
• Google’s AI features in Search: AI Overviews and AI Mode that sit above or alongside web results.

Each asks for different work. Keep them separate in your roadmap and budget.

Use AI to speed up the boring parts. Keep judgement and facts with you. Your site wins on accuracy, structure and offers, not on word count.

• Research: ask for query variants, questions, and page outlines. Then verify in Search Console and a keyword tool you already use.
• Briefs: generate H2s that match search intent and GEO needs. Add the figures, screenshots and examples you will source.
• Drafts: get a first pass for a page like /guides/getting-started. Replace generic claims with your product, data and links.
• Audits: ask for a checklist for /pricing, /blog, and your top ten pages. Then check with your crawler and Search Console.
• QA: compare your draft to the top three results for “{your main query}”. Ensure your page answers the question in the first paragraphs.

Where AI falls short: facts, maths, and your product. It will invent. Give it bullets and data, not a blank page. Keep your style guide short and paste it in each time.

A fixed page looks like this: the H1 says the query, the first 100 words give the answer with a figure or a step, the rest supports it with headings, lists, a table and a date.

GEO means being cited in answers from assistants that search the web. The term came from a 2023 paper by Aggarwal and colleagues. Treat it like a channel with its own work and tracking.

• Where answers come from: assistants retrieve from a web index, then write and cite. Bing’s index powers Copilot and often ChatGPT search. Others crawl on their own too.
• What pages they like: a clear answer near the top, headings that match the question, lists, comparisons, consistent product and people names, quotable figures, a visible date and an author.
• How to be picked: be on the shortlists they already cite. Get on roundups, comparison pages and resource lists that appear in their sources panels.
• How to measure: ask assistants the buyer questions you care about. Log who they cite. Watch referrers like chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com in analytics.

There is no submit button. Assistants cannot be paid for a citation. Your job is to publish pages they can quote and to earn mentions on the pages they already trust.

Set robots.txt with intent. Some crawlers train models. Others power search and answers. Blocking the answering and search crawlers makes you invisible to assistants.

Vendors honour robots.txt voluntarily. Blocking a training crawler does not remove pages already collected. Review your rules twice a year. Keep a comment at the top that states your intent.

# robots.txt example for a small product site
# Goal: allow search and answering crawlers, restrict model training crawlers
User-agent: *
Disallow:

# OpenAI
User-agent: GPTBot
Disallow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /

# Anthropic
User-agent: ClaudeBot
Disallow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /

# Perplexity
User-agent: PerplexityBot
Allow: /
User-agent: Perplexity-User
Allow: /

# Google
User-agent: Googlebot
Allow: /
# Google-Extended controls Gemini training and grounding, not Search
User-agent: Google-Extended
Disallow: /

# Microsoft
User-agent: Bingbot
Allow: /

# Others often seen
User-agent: CCBot
Disallow: /
User-agent: meta-externalagent
Disallow: /
User-agent: meta-externalfetcher
Disallow: /
User-agent: Bytespider
Disallow: /
User-agent: Amazonbot
Disallow: /
User-agent: cohere-ai
Disallow: /

• State the answer in the first 3 to 5 sentences. Example: on /pricing, write “Plans start monthly, cancel any time. All plans include SSO and audit logs.”
• Match the question in H1 and H2s. If buyers ask “{tool} vs {tool}”, your H1 should be “{tool} vs {tool}”. Use the exact names.
• Use lists and tables for facts. Put features and limits in a table. Assistants can quote rows cleanly.
• Keep names consistent. Use the same product and person names across pages, nav and schema.
• Add a visible date and an author. Update the date only when you change the facts.
• Do not hide content behind JavaScript. As of 2026, some crawlers do not execute scripts. Server render the first answer and your key lists.
• Link to sources you rely on. Assistants and editors favour pages that cite.
• Include comparisons and figures they can quote. “The free tier allows 3 projects” is better than “a generous free plan”.
• Target the pages they already cite. Ask “best {category} tools” in Perplexity. If it cites a roundup, get listed on that page.

A fixed comparison page example: /alternatives/your-tool. H1 says “Your Tool alternatives”. First paragraph names the use case, then a shortlist with one-sentence reasons, then a table with features and hints on who each suits.

AI Overviews launched in 2024 and reached many countries in 2024 and 2025. AI Mode, a tab in Search, opened to US users in May 2025 and rolled out wider in 2025. Both use links from Google’s index. There is no separate crawler or opt-out beyond snippet controls.

• Overviews appear on some queries, often questions and informational searches, above the organic results. They cite a handful of pages with links.
• AI Mode answers in prose using a query fan-out and Gemini models. It cites pages ranking in Google already.
• Clicks from both appear in Search Console alongside web search clicks. They are not broken out.
• Your path in: earn organic rankings on the question and comparison queries you target. Get your snippet ready to quote and your page ready to click.

Treat an overview like a featured snippet with friends. Tighten your first paragraph, add a list that fits on one screen, and make your title and meta earn the click.

Know how each assistant finds and cites sources. Then shape your site and outreach to fit.

• ChatGPT search: available to all since December 2024, without an account since February 2025. It mixes Bing and OAI-SearchBot’s own crawl. It shows inline citations and a Sources panel, and it has shopping cards. OAI-SearchBot does not execute JavaScript, so server render what matters.
• Perplexity: searches on every question. It prefers pages that answer in the first paragraphs, with clear headings and recent dates. It shows numbered citations and a sources list. It has Pro Search and a publisher programme.
• Claude: added web search in March 2025. It indexes with Claude-SearchBot and fetches with Claude-User. Answers carry inline citations.
• Gemini: the app answers and, when it searches, grounds with Google Search and shows links. Google-Extended controls Gemini training and grounding in Gemini apps and Vertex AI, not Search.
• Microsoft Copilot: uses Bing’s index. Use Bing Webmaster Tools and IndexNow as usual. Bingbot is the crawler.

• Buy tools that cut cycle time: drafting aids, outline generators, internal link suggestions, schema generators. Keep your editor in charge of facts and tone.
• Buy a crawler that can render and a simple monitor for GEO mentions if GEO is a target channel. You can also track manually on a schedule.
• Avoid any “AI SEO service” that sells thousands of pages or promises AI citations. There is no submission mechanism and assistants cannot be paid for a mention.
• Avoid blocking the wrong bots. Allow Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot if GEO matters. Use Google-Extended and Applebot-Extended if you want to opt out of training.
• Do not buy links on junk lists. Aim to be on the roundups and comparison pages that assistants already cite for your category.

Keep spend small until you prove impact. One good page with real data will beat a hundred spun posts. Your goal is citations, rankings and signups, not volume.

1. 1

   List your buyer questions

   Write 20 to 50 real queries buyers use. Example: “{your tool} alternatives”, “{category} pricing”, “how to {job-to-be-done} with {your tool}”.

2. 2

   Check the assistants weekly

   Ask each question in ChatGPT search, Perplexity, Claude, Gemini and Copilot. Log who is cited, with links. Note if your domain is named.

3. 3

   Watch referrers in analytics

   Segment visits with referrers like chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Track signups from these segments.

4. 4

   Track your target shortlist pages

   List the roundups and comparisons assistants cite for your category. Get listed. Recheck whether those pages then appear in answers that name you.

5. 5

   Review Search Console

   Use the Performance report to see clicks and queries. AI Overviews and AI Mode clicks appear there with web search. Watch question queries for movement.

You can automate some checks with tracking products. They ask assistants at scale and log mentions. Still, keep a small manual set you read yourself each week to learn what changed and why.

As of 2026, Google judges AI-written content like any content. It acts on pages made at scale without value of their own. The helpful content system signals were folded into core ranking in March 2024.

• Avoid scaled content abuse: mass pages to manipulate rankings. Your fix is to publish fewer pages, each with data or an answer of its own.
• Avoid site reputation abuse: do not host third-party content to borrow your domain’s trust.
• Do not revive expired domains for their old links. Build your own pages and mentions.
• Show your data. Screenshots, metrics, and product steps are signals of value that AI alone cannot fake.

A safe scaled play looks like this: a programmatic set of city or use-case pages where each page carries unique data, steps, screenshots and internal links that route to support and pricing. If you cannot add that, do not scale yet.