Tag: ChatGPT SEO

By Faisal Abbasi, CEO of Audit & Rank · Updated October 14, 2026 · 14 min read

Most articles on this topic stop at “write helpful content and add FAQ schema.” That advice is not wrong, but it does not explain why one page gets cited and an equally good page does not. This guide covers the mechanics: how AI search systems pick passages, which crawlers you must let in, how to write paragraphs a model can quote, how to build the off-site signals that drive recommendations, and how to measure any of it.

AEO vs GEO vs SEO: what is the difference?

The practical takeaway: SEO gets you into the pool of candidate sources. AEO and GEO decide whether you get picked from that pool.

How AI search actually chooses its sources

The exact systems differ, but the major AI search products follow the same broad pipeline. Knowing the steps tells you where you can win.

1. The question is expanded into many searches

Google describes AI Mode as using “query fan-out”: one question is split into several related sub-queries that run at the same time. A user asking “is a heat pump worth it in Ohio” may trigger hidden searches about running costs, cold-climate performance, rebates and installation price. Each sub-query is a separate chance for your page to be retrieved. Pages that only target the head keyword miss most of these.

2. Candidate pages are retrieved from an index

Google’s AI features draw on Google’s own search index, so normal indexing rules apply. ChatGPT search relies on OpenAI’s search crawler, OAI-SearchBot, alongside its search partners. Perplexity runs its own crawler. If the crawler for a given engine is blocked, you are invisible on that engine no matter how good the content is.

3. Passages, not pages, are scored

The model does not read your whole article. It works with chunks: a heading and the paragraphs under it, a table, a list. A chunk that answers a sub-query clearly and on its own is far more likely to be used than a great answer buried in paragraph nine of a story.

4. The answer is written and sources are cited

The model writes a response and attaches citations to the passages it leaned on. For “best” and “which company” questions, it also weighs what other sites say about each brand. That is where reviews, directories and third-party mentions come in.

The five problems we see most in client audits

These come from real audit conversations, not keyword tools. If one sounds familiar, the fix is in the sections below.

“Our rankings did not drop, but clicks did.”

This is usually an AI Overview sitting above your result and answering the question before anyone scrolls. You keep the impression and lose the click. The response is to target the queries where the click still has value (pricing, comparisons, local intent, tools) and to make sure your brand is the one cited inside the overview.

“ChatGPT recommends our competitor, never us.”

Check two things. First, whether OAI-SearchBot can reach your site at all; firewalls and CDN bot settings block it more often than robots.txt does. Second, how often your brand appears on third-party pages that discuss your category. Language models recommend the brands that the web already talks about.

“We added FAQ schema and nothing happened.”

Schema helps machines understand what a page contains. It does not make weak content citable. Since 2023 Google shows FAQ rich results mainly for authoritative government and health sites, so the visible snippet benefit for most businesses is gone. Keep the markup for clarity; fix the content first.

“Do we need an llms.txt file?”

llms.txt is a proposed file that lists your key pages for language models. Google has said it does not use it for Search, and no major AI search engine has confirmed it as a ranking or citation signal. It takes ten minutes to add, so add it if you like, but do not expect results from it.

“How do we even know if AI search sends us anything?”

Most sites never separate AI referrals in analytics, so the traffic hides inside “referral.” The measurement section shows how to fix that in an afternoon.

Layer 1: technical access for AI crawlers

Each AI company uses different crawlers for different jobs. Blocking the wrong one removes you from search; allowing the wrong one may let your content be used for training when you did not want that. Decide per bot.

A sensible default for a business that wants AI visibility but not training use:

User-agent: Googlebot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Bingbot
Allow: /

User-agent: GPTBot
Disallow: /

Sitemap: https://www.example.com/sitemap_index.xml

Three checks most people skip:

• Your CDN or firewall. Some CDNs now block AI crawlers by default on new sites, and security plugins often rate-limit unknown bots. robots.txt can say “Allow” while the firewall returns a 403. Check the bot or WAF settings in your hosting and CDN dashboards, then look at server logs for successful requests from these user agents.
• JavaScript rendering. Most AI crawlers fetch the raw HTML and do not reliably run JavaScript. If your prices, service list or FAQ only appear after a script loads, many AI systems never see them. View the page source (not the inspected DOM) and confirm your key text is there.
• Snippet controls. Google’s AI features only use pages that are indexed and allowed to show a snippet. A stray nosnippet or a very low max-snippet value removes you from AI Overviews while you still rank normally.

Structured data that is worth the effort

• Organization with sameAs links to your LinkedIn, Google Business Profile and other official profiles, so engines connect your brand as one entity.
• Person markup for authors, linked from Article, so expertise is attached to a real name.
• LocalBusiness (or a subtype like HVACBusiness or Dentist) with address, area served and opening hours for any local service.
• Product or Service with price ranges where you can publish them. AI answers to “how much does X cost” favour pages with actual numbers.

Layer 2: writing passages an AI can quote

This is where AEO lives. The aim is simple: every section should contain at least one passage that answers a specific question completely, in a form a model can lift without rewriting.

The answer-first pattern

1. Phrase the heading as the question people ask, or close to it.
2. Answer it in the first 40 to 60 words under the heading, in plain language, with the key number or fact.
3. Then add the detail: conditions, exceptions, examples, sources.

Here is the difference on a typical local service page:

The second version answers the question, contains numbers and named parts, carries a year, and makes sense if a model pulls it out alone. The first version contains nothing a model can use.

Rules that make passages self-contained

• Avoid “as mentioned above” and “this” without a noun. A lifted passage has no “above.”
• Name things fully: “Google Business Profile,” not “your profile.”
• Use units, ranges, dates and locations. Vague claims are not quotable.
• Put comparisons in tables. Models extract tables cleanly.
• Keep one idea per paragraph so chunk boundaries fall in sensible places.

Add what only you know

AI answers are built from the consensus of many pages. If your page says what twenty others say, the model has no reason to cite you in particular. Pages that get cited usually add something new: your own pricing data, results from real projects, a test you ran, a local detail, a photo of the actual problem. Research on generative engines from Princeton and partners (Aggarwal et al., 2024) found that adding statistics, quotations and citations to sources raised a page’s visibility in AI-generated answers by up to about 40% in their tests. Specific evidence beats polished prose.

Show who wrote it and when

Put a named author with relevant experience on every article, link to an author page, and show a real “updated” date when you change something meaningful. For pricing, laws and tools, freshness changes which source looks safest to quote.

Layer 3: the entity and reputation layer (where GEO is won)

When someone asks “who is the best SEO agency for dentists” or “best foundation repair company in San Antonio,” the model is not reading one page. It is summarizing what the web broadly says. To appear in that summary, your brand has to be part of the conversation off your own site.

• Reviews where AI looks. Google Business Profile, plus the platforms for your field: Clutch or G2 for agencies and software, Yelp and Angi for home services, Avvo for lawyers. Volume and recency both matter.
• Consistent business details. The same name, address, phone and description everywhere. Inconsistent data splits your brand into several weaker entities.
• Category mentions. Get included in “best of” lists, local news, industry roundups and partner pages. These are the pages that “best X” prompts draw on.
• Community discussions. Reddit and niche forums are frequently cited by AI answers. Useful, non-promotional answers from a real account build mentions over time.
• Wikidata. For established brands with independent coverage, an accurate Wikidata entry strengthens entity recognition.

How to measure AI search visibility

Google Search Console

Clicks and impressions from AI Overviews and AI Mode are counted inside your normal Performance report. On June 3, 2026, Google announced dedicated Generative AI performance reports that break this out, showing impressions by page, country, device and date (with data from May 18, 2026). At launch they reached only a subset of UK sites, with a wider rollout planned, and they do not yet show clicks, CTR or queries. Check whether your property has them; if not, watch for the rollout.

Google Analytics 4

Create a custom channel group so AI referrals stop hiding inside “Referral.” Add a channel named “AI Assistants” above Referral, with the condition “Source matches regex”:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai

Expect the numbers to be small at first and conversion rates to be higher than average; people arriving from an AI answer have usually already compared options.

A monthly prompt panel

No tool sees every AI answer, so build your own sample. Write 20 to 30 prompts your buyers would really type, such as “best emergency plumber in Tampa” or “how much does a dental implant cost in Dallas.” Once a month, run them in ChatGPT, Perplexity, Gemini and Google AI Mode and record for each one: cited with a link, mentioned without a link, or absent. The trend over three months tells you more than any single check.

A 30-day plan you can actually follow

1. Week 1, access. Review robots.txt, CDN and firewall bot settings, and snippet tags. Confirm in server logs that Googlebot, OAI-SearchBot and PerplexityBot get 200 responses. Make sure key content is in the raw HTML.
2. Week 2, top pages. Pick the ten pages that bring leads or sales. Rewrite each section answer-first, add real numbers, a comparison table where it fits, an author and an updated date.
3. Week 3, entity. Clean up Organization and LocalBusiness schema with sameAs, fix inconsistent listings, and start a steady review request routine.
4. Week 4, measure. Set up the GA4 AI channel, check Search Console, and run your first prompt panel as the baseline.

Want all four weeks checked for you? Our SEO audit covers crawler access, rendering, schema, top pages and AI visibility in one report.

Frequently asked questions