The short answer: you do not "rank number one" in ChatGPT in the same stable way you can rank first in a traditional search result. AI answers are generated dynamically, can vary between runs, and may use different sources depending on the prompt, location, freshness needs and whether live web search is triggered.
The practical goal is broader: become a source and brand that AI systems can find, retrieve, understand, cite and recommend when people ask questions connected to your market.
That still starts with SEO. But it also requires clear entity information, genuinely useful source material, strong third-party corroboration and a different measurement model. A single "ChatGPT ranking" checked once is not enough.
This guide separates what OpenAI, Google, Anthropic and other platforms actually document from what the SEO industry has inferred through testing. Where evidence is limited, I say so.
What does "ranking in ChatGPT" actually mean?
The phrase is useful because it matches how people search, but it can create the wrong mental model. ChatGPT does not normally present a fixed ten-result ladder where one page permanently occupies position one.
In practice, AI visibility can mean several different things:
Your page is cited
A ChatGPT answer links directly to your page as supporting evidence.
Your brand is mentioned
The answer names your company, product or expert even when it does not link to you.
Your brand is recommended
You appear in a shortlist, comparison or suggested solution for a relevant need.
Your information is used
Your facts, definitions, research or examples shape the generated answer, whether attribution is visible or not.
Those outcomes are related but they are not identical. A site can be frequently cited without being recommended as a brand. A business can be recommended because third-party sources talk about it even when its own website is not the cited source.
This is why I prefer the term AI search visibility when discussing the whole problem. "ChatGPT SEO", "LLM SEO", "GEO" and "AEO" are useful industry labels, but the job is fundamentally about increasing accurate visibility across generated answers.
Stop asking only "What position am I?" Start asking: "For which customer questions am I retrieved, cited, mentioned and recommended?"
How ChatGPT finds and cites information
ChatGPT can answer from model knowledge, use live web search, or combine multiple information sources depending on the request and product experience. The part website owners can influence most directly is the live retrieval layer.
OpenAI's current publisher guidance says any public website can appear in ChatGPT search and specifically tells publishers not to block OAI-SearchBot if they want content to be discovered, surfaced, cited and linked. OpenAI also says ChatGPT may obtain a URL through a third-party search provider or through crawling other pages.
That last point is important. It means you should not build your strategy around the claim that "ChatGPT uses only Bing" or "ChatGPT uses only Google". OpenAI does not publicly document one exclusive search index for all ChatGPT search experiences.
OAI-SearchBot and GPTBot are not the same thing
OAI-SearchBot
Used for website discovery and ChatGPT search visibility. If organic ChatGPT search visibility matters to you, this is the crawler OpenAI explicitly tells publishers to allow.
GPTBot
Used in relation to model training. OpenAI documents this separately. You can make a different decision about training access without treating it as the same question as search visibility.
OpenAI also says ChatGPT search ranking uses multiple factors intended to help users find relevant, reliable information. It does not publish the formula or guarantee placement.
Retrieval happens before citation
Industry research helps explain why eligibility alone is not enough. Ahrefs analysed a large set of ChatGPT prompts and found that ChatGPT retrieves many URLs it never ultimately cites. Their 2026 study found strong differences between retrieval channels and showed that semantic relevance to the behind-the-scenes sub-questions was important within the search retrieval set.
That suggests a useful distinction:
- Discovery: can the system find the URL?
- Retrieval: is the URL selected as a candidate for this prompt?
- Citation: does the final answer actually credit it?
- Recommendation: does the generated response positively surface the brand or solution?
Those are different optimisation problems.
The AI Visibility Funnel
I use a five-stage model because it prevents teams jumping straight to content formatting while ignoring more fundamental problems.
Your page is publicly accessible, crawlable and not blocked by the relevant search or AI crawler.
The page is relevant and authoritative enough to enter the candidate set for a prompt or fan-out query.
The content contains specific, defensible information the system can use and attribute cleanly.
Your wider reputation and evidence support the idea that your brand belongs in a recommendation or comparison.
You track mentions, citations, sentiment, referral visits and commercial outcomes across repeated prompts over time.
1. Make your site eligible for AI search
Technical eligibility is the boring part, but it is the right place to start because it is binary. If the relevant crawler cannot access your important pages, nothing further down the funnel matters.
Check OAI-SearchBot first
OpenAI specifically recommends allowing OAI-SearchBot. A simple robots.txt rule can look like this:
You do not need to add an explicit allow rule if your robots.txt already permits crawling. The real objective is to make sure you have not blocked it accidentally.
Check your CDN, firewall and bot protection
Robots.txt is only one layer. OpenAI's current crawler guidance warns that CDNs and bot-protection systems can return 403 errors or challenge legitimate crawlers. If you use Cloudflare, Akamai or another security layer, inspect bot rules and server logs rather than assuming robots.txt tells the whole story.
OpenAI publishes searchbot IP information for verification. Use verified bot signals where available rather than blindly trusting a user-agent string, which can be spoofed.
Do the same for other platforms you care about
| Platform | Search / retrieval access | What to know |
|---|---|---|
| ChatGPT | OAI-SearchBot | OpenAI explicitly links this crawler to discovery and search visibility. |
| Google AI Overviews / AI Mode | Google Search indexing | Google says pages must be indexed and eligible for a normal Search snippet. No separate AI crawler or special AI file is required. |
| Claude | Claude-SearchBot / Claude-User | Anthropic separates search indexing from user-directed retrieval and model-training crawling. |
| Perplexity | PerplexityBot | Perplexity documents its own search crawler and maintains its own large-scale search index. |
| Microsoft Copilot | Bing search ecosystem | Microsoft documents Copilot in Bing as being grounded in web search results and high-authority web content. |
Do not confuse training access with search access
This distinction is particularly important for OpenAI and Anthropic. A publisher may have one preference for model-training crawlers and another for live search or user-directed retrieval. Treat those choices separately.
2. Win the retrieval layer with strong SEO fundamentals
AI search has not made traditional SEO irrelevant. If anything, the strongest platform documentation points in the opposite direction.
Google says its generative AI features are rooted in its core Search ranking and quality systems. To appear as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to appear in Google Search with a snippet.
ChatGPT is less transparent about its ranking mechanics, but OpenAI says results are ranked using multiple factors for relevance and reliability. Large-scale industry research also suggests general web-search retrieval is a major route into ChatGPT citations.
What still matters
- Clear crawl paths and internal links
- Unique, indexable URLs
- Titles that accurately describe the page
- Content that genuinely matches the query or problem
- Strong topical context across the site
- Useful external links and citations where evidence matters
- Real authority and reputation, including relevant links and mentions
- Fast, usable pages for the humans who actually visit
I would not create a separate "AI SEO site" detached from your normal SEO architecture. The better approach is to make your existing site more useful, explicit and evidence-rich.
Do titles still matter?
Yes, but not because there is a secret ChatGPT title formula. Ahrefs' 2026 citation research found that title and content relevance to the sub-questions ChatGPT generates behind the scenes can influence which retrieved pages are ultimately cited. That is another argument for accurate, specific titles rather than clever but vague headlines.
3. Optimise for real prompts and the questions behind them
Traditional keyword research usually starts with a short query such as "best accounting software". Conversational search expands that into much richer requests:
"I run a five-person UK agency, use Xero, need project profitability reporting and do not want an enterprise contract. Which accounting system should I shortlist and why?"
The page that answers only the head term may not satisfy the actual decision.
Think in prompt families, not one magic prompt
For each commercial topic, map the different ways a real customer can frame the problem:
- Discovery: What options exist?
- Comparison: X vs Y, or best options for a specific situation
- Qualification: Which option is suitable for this industry, budget or constraint?
- Risk: What can go wrong? What should I avoid?
- Implementation: How do I do this properly?
- Evidence: What data, examples or case studies support the recommendation?
Then look at the sources the platforms actually cite for those prompt families. Sometimes the opportunity is your own guide. Sometimes the more important opportunity is getting your brand included in a respected third-party comparison or review.
Fan-out does not justify hundreds of thin pages
AI search systems can decompose a complex prompt into sub-questions. That does not mean you should create a separate page for every sentence somebody might type. Google's 2026 AI-search guidance explicitly warns against scaled content created primarily to capture every query variation.
A strong page can cover a coherent subject deeply. Split content when the intent or task genuinely deserves its own page, not because a tool produced 500 long-tail prompts.
4. Publish information worth citing
This is where most generic "GEO" advice becomes too mechanical. A model does not need another paraphrase of information already available from ten stronger sources.
Google's current generative-AI guidance puts unusual emphasis on unique, valuable, non-commodity content. That is a useful principle across AI search more broadly, even though each platform works differently.
Give the web something it did not already have
Original research
Your own dataset, survey, experiment, benchmark or analysis, with methodology and limitations.
First-hand experience
What happened when you actually implemented, tested, bought, fixed or compared something.
Useful comparisons
Specific trade-offs, suitability criteria and constraints instead of generic pros and cons.
Distinct analysis
A defensible framework or conclusion that helps the reader interpret the evidence rather than merely summarise it.
Citable content is specific
Definitions, numbers, named examples, procedures and clear comparisons give an answer engine something concrete to use. Vague claims such as "AI is transforming marketing faster than ever" are easy to generate and difficult to justify.
If you quote a statistic, link to the primary source where possible. If it is your own number, explain where it came from. If it is an opinion, write it as an opinion.
The original GEO research is useful, but easy to overstate
The academic paper that popularised the term Generative Engine Optimization found visibility gains from techniques including citations, quotations and statistics in its experimental setup. That does not prove a universal ChatGPT ranking recipe. Later research has stressed that effects vary by domain and that the GEO problem spans several stages, including retrieval, citation and answer use.
Use research as evidence, not as a licence to manufacture fake quotations or bolt arbitrary statistics onto every paragraph.
5. Make the content easy to understand and extract
Clear structure helps humans first. It also gives search and AI systems cleaner relationships between questions, answers, lists, comparisons and evidence.
I would optimise for answer clarity, not arbitrary "LLM chunks".
Use direct answers where the question deserves one
If a heading asks "Does ChatGPT use Bing?", answer that question immediately, then add the nuance. Do not make the reader wade through 400 words of scene-setting to discover the answer.
That does not mean every answer must be 40 words or 80 words. There is no official OpenAI rule specifying an ideal citation-block length, and Google explicitly says there is no required content chunk size for generative search.
Use the right HTML element for the information
- Headings for genuine topic hierarchy
- Lists for real sets of items or steps
- Tables for genuine comparisons and relationships
- Links that describe where they go
- Figures and captions when visuals carry evidence
- Accessible form controls and interactions
This overlaps with the semantic alignment framework I use for SEO and AI search: content, HTML, visual design, structured data, media and interaction should describe the same thing.
Structured data helps clarity, but there is no ChatGPT schema
Use normal schema when it truthfully describes the page and supports standard search features. Article, Person, Organisation, Product, LocalBusiness and other appropriate types can clarify entities and properties.
Do not invent special "AI schema". Google says no special schema is needed for its generative AI features, and OpenAI does not publish a ChatGPT-specific schema requirement.
6. Make your entities and trust signals unambiguous
AI answers often need to decide not just whether a fact is relevant, but who or what the fact belongs to.
Your website should make core entities easy to identify:
- Who operates the business?
- What products or services does it actually provide?
- Where does it operate?
- Who wrote or reviewed important content?
- What experience or qualifications support the claims?
- What evidence exists outside the website?
Think E-E-A-T, but do not turn it into a fake score
Google's E-E-A-T framework is useful because it asks whether experience, expertise, authority and especially trust are visible and supportable. It is not an official ChatGPT ranking system, but the underlying credibility problem exists on every answer engine.
Real evidence can include first-hand examples, author expertise, case studies, primary-source citations, clear ownership, transparent policies and independent reputation. If you want the fuller framework, see my guide to E-E-A-T in SEO.
Keep the core brand story consistent
Your website, professional profiles, key directories, review platforms and third-party mentions should not describe completely different businesses. Consistency here means factual consistency, not copy-pasting the same sentence everywhere.
A brand can offer several services and be described in different words. The problem is contradiction: different names, outdated locations, conflicting product claims or incompatible descriptions of what the company actually does.
7. Build third-party corroboration, not manufactured mentions
If the only place saying you are the best solution is your own website, an answer engine has limited independent evidence.
Third-party sources can matter in two different ways:
They can be cited directly
A review site, industry publication, comparison page or news source may be the page an AI system actually cites when discussing your brand.
They can corroborate the entity
Independent mentions help establish that the business, product, expert or claim exists beyond your own marketing copy.
Useful ways to earn that corroboration
- Digital PR built around real stories, data or expertise
- Relevant industry coverage and interviews
- Genuine customer reviews on platforms people actually use in your sector
- Inclusion in editorial comparisons where you genuinely belong
- Expert contributions to credible publications
- Original research other writers want to cite
- Useful participation in communities when you actually have something to add
Reddit and YouTube need nuance
Some studies show AI systems retrieve large amounts of community and video content. That does not mean "spam Reddit" or "make YouTube videos because ChatGPT always cites them". Ahrefs' 2026 analysis, for example, found very low direct citation rates for some dedicated Reddit and YouTube retrieval channels, even though those sources appeared to contribute context.
8. Keep important content fresh without faking freshness
AI search often handles current questions, so freshness matters when freshness is relevant. It matters much more for software features, prices, regulations, news, statistics and rapidly changing recommendations than for stable evergreen facts.
Research on ChatGPT citations has found a broad freshness skew compared with traditional search results, but the same research also shows that relevance still does the heavy lifting. Simply changing a date does not make an irrelevant page more useful.
A meaningful update should change the content
- Recheck claims against current primary sources
- Replace outdated screenshots and examples
- Update product capabilities, prices or policies
- Add new first-hand evidence or research
- Remove advice that is no longer defensible
- Update dateModified only when the page was genuinely revised
For a guide like this one, the platform-specific sections deserve regular review because crawler names, measurement tools and search behaviour change quickly.
ChatGPT vs Google AI vs Perplexity vs Claude vs Copilot
The platforms overlap, but they are not one ranking system. Build one strong foundation, then account for the differences that are actually documented.
ChatGPT
Allow OAI-SearchBot, keep your pages reachable through the host/CDN, and focus on relevance and reliability. OpenAI says placement is not guaranteed and may use third-party search providers alongside its own crawling.
Google AI Overviews & AI Mode
Google says normal SEO remains the foundation. Pages need to be indexed and snippet-eligible. No special AI schema, llms.txt or artificial chunking is required.
Perplexity
Perplexity operates its own search infrastructure and documents PerplexityBot. Freshness and document understanding are explicit parts of its indexing architecture.
Claude
Anthropic distinguishes Claude-SearchBot, Claude-User and ClaudeBot. Search visibility and user-directed retrieval can therefore be managed separately from model-training crawling.
Microsoft Copilot
Microsoft has documented Bing Copilot as being grounded in web search results, with Bing search quality, relevance, credibility and freshness forming part of that retrieval environment.
Cross-platform reality
Expect source overlap to be imperfect. Different engines, prompt wording and repeated runs can produce different sources. Optimise for broad credibility and test each platform separately.
Local businesses and ecommerce sites need extra data layers
For some queries, an AI system does not rely only on ordinary editorial webpages. Business listings, product feeds, reviews, maps and structured merchant data can become part of the answer.
Local businesses
Keep your Google Business Profile accurate because Google's own generative-search guidance specifically points local businesses towards Business Profile data. Maintain real business information across the places customers use, and make sure the website itself clearly explains services, locations and evidence.
For ChatGPT and other assistants, local visibility can also depend on third-party sources and current location-aware search results. Do not assume one local directory controls all AI recommendations.
Ecommerce
Accurate product data, availability, price, images, reviews and merchant feeds can matter as much as editorial copy. Google's guidance explicitly recommends Merchant Center for product visibility in its AI experiences. OpenAI's current crawler guidance also references product feeds and product images in its commerce workflows.
For product pages, I would prioritise:
- Unambiguous product and variant names
- Current price and availability
- Accurate Product structured data
- Specific specs and comparison data
- Original product images and useful media
- Real review and testing evidence
- Clear shipping, returns and support information
AI SEO tactics I would not rely on
This category attracts confident advice because the systems are opaque and changing quickly. That makes evidence discipline particularly important.
| Tactic or claim | My view | Why |
|---|---|---|
| "Rank #1 in ChatGPT" | Misleading | Generated answers vary. Visibility should be measured across repeated prompts, citations, mentions and recommendations. |
| llms.txt is required | Unproven | OpenAI does not document it as a ChatGPT Search requirement. Google explicitly says it ignores llms.txt for Search and AI features. |
| FAQ schema boosts AI citations | Unsupported | There is no documented ChatGPT citation benefit and Google says no special schema is needed for generative search. |
| Every answer must be 40-80 words | Unsupported | Clear answers help, but no major platform publishes a magic passage length. |
| Publish hundreds of thin pages | High risk | Google warns against scaled content and pages made mainly to capture query variations. |
| Manufacture Reddit or forum mentions | Bad strategy | Community data may influence context, but fake consensus is fragile and can violate platform or search-spam rules. |
| Change the date every month | Do not do it | Freshness is useful when the information actually changes. Cosmetic date changes do not improve relevance. |
| Traditional SEO no longer matters | Wrong | Google explicitly says SEO remains foundational, and web-search retrieval is a major route into AI answers. |
How to measure ChatGPT and LLM visibility properly
AI visibility is probabilistic. One prompt, run once, is a screenshot, not a measurement system.
Track four different outcomes
Brand mentions
How often is the brand named for important prompts, whether or not the website is linked?
Citation share
Which domains and pages are cited, and what percentage of repeated runs include yours?
Context and sentiment
Are you recommended, neutrally described, excluded for a reason, or framed negatively?
Business outcome
Referral visits, assisted conversions, leads, branded search and direct visits influenced by AI discovery.
Use repeated tests and prompt variants
A sensible manual benchmark might take a defined set of commercially relevant prompts, run them repeatedly, and repeat the exercise monthly using the same methodology. Include realistic paraphrases because AI search can be sensitive to wording.
For a serious programme, track:
- Prompt family
- Platform
- Date and location context where relevant
- Whether web search was used
- Brands mentioned
- URLs cited
- Recommendation order or framing
- Reason given for inclusion or exclusion
Measure actual traffic too
OpenAI says ChatGPT search referrals include utm_source=chatgpt.com, so you can segment that traffic in analytics. Referral traffic will not capture every influence because users may search your brand later, visit directly or act without clicking.
For Google's AI experiences, use the dedicated Generative AI performance reporting in Search Console where available. Do not mix all AI platforms into one metric and assume they behave identically.
A practical 90-day AI search plan
If I were starting from zero, I would prioritise the work in this order.
Days 1-30: eligibility and baseline
- Check OAI-SearchBot and other relevant crawlers
- Inspect CDN/WAF blocks and HTTP responses
- Confirm Google and Bing index health
- Define 20-50 commercially meaningful prompt families
- Record current citations, mentions and competitors
- Identify which pages and third-party sources are already winning
Days 31-60: improve the candidate set
- Fix weak title and intent alignment
- Upgrade thin or generic pages with first-hand evidence
- Add direct answers, comparisons and specific data
- Strengthen author, organisation and service/product clarity
- Improve internal linking around core topics
- Update stale high-value content
Days 61-90: build corroboration and measure
- Earn relevant third-party coverage and references
- Pitch original data, commentary or case studies
- Strengthen genuine review and reputation signals
- Re-run the prompt benchmark with the same method
- Compare citation share, mention share and context
- Prioritise the next cycle from actual gaps, not generic checklists
The final framework: optimise the evidence chain
The best way to approach ChatGPT SEO and LLM visibility is not to hunt for a new list of secret ranking factors. Build a stronger evidence chain from discovery to recommendation.
The AI search visibility checklist
- Eligibility: the right crawlers and search engines can reach your content.
- Retrieval: your pages are relevant, indexable and competitive for the questions people actually ask.
- Information gain: you publish something worth selecting over another summary.
- Clarity: definitions, facts, comparisons and evidence are easy to understand in context.
- Entity confidence: systems can tell who you are, what you do and which claims belong to you.
- Corroboration: independent sources support important brand and expertise claims.
- Freshness: changing information is kept accurate without fake date updates.
- Measurement: you track repeated prompts, citations, mentions, context, referrals and conversions.
If that sounds a lot like good SEO, good content, digital PR and brand building working together, that is because it is.
AI search changes the output. Instead of competing only for a blue-link position, you are competing to become part of the evidence a system uses to construct an answer.
The durable strategy is not to make content look "AI optimised". It is to become easy to find, easy to verify and worth citing.
Common questions about ranking in ChatGPT and AI search
Sources and further reading
- OpenAI Help Center - Searching the web with ChatGPT
- OpenAI Help Center - Publishers and Developers FAQ
- OpenAI Help Center - Guidance for allowing OpenAI web crawlers
- Google Search Central - Optimizing your website for generative AI features
- Google Search Central - AI features and your website
- Google Search Central - Generative AI performance reports in Search Console
- Anthropic Help Center - Claude web crawlers
- Anthropic Help Center - How Claude web search works
- Perplexity - Architecting and evaluating an AI-first search API
- Microsoft - Bing Systemic Risk Assessment and grounded Copilot search
- Ahrefs - How to Rank on ChatGPT: What Actually Works
- Ahrefs - Why ChatGPT Cites One Page Over Another, study of 1.4M prompts
- Semrush - How to rank in ChatGPT search
- Aggarwal et al. - GEO: Generative Engine Optimization, KDD 2024
- Martinez - Critical Survey of Generative Engine Optimization, 2026
- Chen et al. - Generative Engine Optimization: How to Dominate AI Search
