Ranking in AI assistants is not one problem. It's four. ChatGPT reads from Bing's index. Claude reads partly from its training data and partly from Brave Search. Perplexity re-crawls the web at query time. Google AI Overviews pulls from Google's own index with new reranking logic. The signals overlap but they are different enough that you cannot optimize for one and hope the rest follow.
This is the practitioner guide to each one, in the order they actually matter by market share: ChatGPT (52% of AI assistant traffic), Gemini and AI Overviews (28%), Claude (9%), Perplexity (1.3% but disproportionately influential because every answer cites sources inline).
Each section tells you the same three things: what the AI reads, what gets cited, and the specific tactics that work.
How to rank in ChatGPT
What ChatGPT reads: ChatGPT Search retrieves live results from Bing's index and then rewrites them with citations. Pages missing from Bing are effectively invisible to ChatGPT.
What gets cited: an AirOps study found ChatGPT cites only 15% of the pages it retrieves, running the retrieved set through a four-gate funnel of Bing index, crawler access, candidate selection, final citation.
Specific tactics:
- Verify your site in Bing Webmaster Tools. Submit an XML sitemap. Activate IndexNow so Bing picks up every publish and edit within minutes instead of weeks.
- Allow OAI-SearchBot in your robots.txt. This is OpenAI's dedicated crawler for ChatGPT Search. Blocking it kills your citations.
- Put a direct answer in the first 100 words of each page, under a question-shaped heading. If a user types the question into ChatGPT, your first paragraph should be the answer the model lifts.
- Add numbers to your claims. A claim with a figure attached gets cited far more than a vague assertion. "Costs around $3,500" beats "affordable pricing" every time.
- Build entity presence on Wikipedia, Wikidata and Crunchbase. These are high-weight sources in the training data that shaped GPT. Brands that rank well in Bing but have no Wikipedia or Reddit presence under-cite in ChatGPT.
- Add FAQPage and Article schema. Both are extractable in a format ChatGPT lifts cleanly.
- Use semantic URLs. Pages with descriptive slugs produce measurably more citations than hash-based or ID-based URLs.
How to rank in Perplexity
What Perplexity reads: Perplexity runs real-time web retrieval at query time. It does not rely on a pre-built index. It re-crawls content when the user asks the question.
What gets cited: Perplexity's ranking factors break down to roughly 30% content relevance, 20% visual placement, 15% domain authority, 15% content freshness, 10% source diversity and 10% structured data. The reranker heavily favors answer-first content and recent dates.
Specific tactics:
- Lead every section with a direct 30-to-60 word definition immediately after the H2. Pages using Q&A or direct answer format hit a 55% top-3 citation rate versus the 31% average.
- Add FAQPage schema. Pages with FAQPage markup are 3.2x more likely to appear in Perplexity responses than equivalent prose.
- Refresh pillar content quarterly. Perplexity's reranker weights recency aggressively. A page that was definitive six months ago gets passed over for a competitor who updated last week.
- Include visible publication dates and "last updated" timestamps. The reranker reads these.
- Add author bylines with linked bios. Credentials, LinkedIn profile, professional website. Perplexity treats byline authority as a top-tier signal.
- Use comparison tables, bulleted lists and clear section headers. These structural formats dominate Perplexity's citation set.
How to rank in Claude
Claude is different. It has two citation layers that work independently.
Layer 1: Claude's default answer mode is training-corpus-only. Citations in that mode depend on entity-level recognition built up across the open web before the training cutoff. Wikipedia and structured reference data are heavily weighted. Entity presence on those sources is the single most important move.
Layer 2: When Claude uses web search, it retrieves results from Brave Search. Brave visibility and freshness drive live citations. Nobody optimizes for Brave. The ones who do win.
Specific tactics:
- Get into Wikipedia. If your business is notable enough for a Wikipedia page, make sure it exists and is clean. If not, get mentions on Wikipedia pages adjacent to your category.
- Get on Wikidata and Crunchbase. These are structured entity sources Claude's training data weighs heavily.
- Submit your site to Brave Search Webmaster Tools. Brave is a smaller index than Bing but it is the exclusive live-search source for Claude. First-mover advantage is massive.
- Diversify your citation surface. The top 10 domains account for just 9.5% of all Claude citations. You need roughly 500 domains to reach 60% of citations. Translation: Claude rewards breadth of mention across the open web, not concentration on your own site.
- Allow Claude-User in robots.txt. That is Anthropic's real-time retrieval bot. You can block ClaudeBot (training crawls) separately if you do not want your content training future Claude models.
- Skip social media as a citation strategy. Social platforms account for just 0.9% of Claude citations. Time spent there does not move the needle for Claude.
How to rank in Google AI Overviews and Gemini
Google AI Overviews are the AI-generated summaries at the top of Google results. Gemini is Google's standalone chat assistant. Both pull from the same Google index with new reranking.
What gets cited: semantic completeness is the number-one ranking factor. Content scoring 8.5 or higher on semantic completeness is 4.2x more likely to be cited. AI Overviews favor passages that fully answer the query in 134 to 167 word self-contained chunks. Pages with 15 or more recognized entities show 4.8x higher selection probability. 96% of AI Overview citations come from sources with strong E-E-A-T signals. Schema-marked pages are cited 2.3x more often.
Specific tactics:
- Write passages that are self-contained. Each 150-word block should answer one question completely without requiring context from surrounding paragraphs. This is how AI Overviews extract and present content.
- Density of recognized entities matters. Named people, companies, products, places and concepts, all with internal links or external references.
- Visible named author with linked bio, publication date and external citations are now mandatory for strong E-E-A-T signaling.
- Add schema. Article, FAQPage, HowTo and Breadcrumb are the highest-impact types.
- Do not assume ranking first on Google means AI Overview citation. Only 38% of cited pages now also appear in the top 10 search results for the same query, down from 76% in mid-2025. AI Overviews run their own reranking that pulls from positions 11-100 and beyond at near-equal rates.
- Create a YouTube channel. YouTube is the single largest source cited in Google AI Overviews, at 20.9% citation share. Structured channels with clear titles, descriptions, uploaded transcripts and chapter markers earn AIO citations at rates uncoupled from the domain authority of their website.
What applies to all four
Six moves work regardless of which AI you are targeting:
- Answer-first content. Direct answer in the first 100 words of each section, under a question-shaped heading.
- FAQPage schema. The one schema type that measurably lifts citation rates on every assistant.
- Entity consistency. Same business name, same description, same services across your site, Google Business, LinkedIn, directories, press, podcasts.
- Visible publication dates and author bylines with real credentials.
- Numbers over adjectives. "$3,500/month" beats "competitively priced."
- Freshness. Content that was definitive last year is being replaced by content that was updated last month.
The four moves that do NOT work the same way across all four:
- Google rank. Still the primary signal for ChatGPT (via Bing overlap) and still useful for AI Overviews, but not meaningful for Claude or Perplexity in the way it used to be.
- Social media. Marginal for Perplexity, invisible for Claude. Do not budget major time here for AI citation.
- Backlinks. Still matter for Google and ChatGPT. Matter less for Perplexity (which has its own authority model) and less again for Claude (which cares about entity breadth more than inbound links to any single page).
- Keyword stuffing. Dead for all four. The 2026 rerankers penalize it directly.
How to prioritize when you only have time for one
If you have one hour this week to improve your AI visibility:
Add FAQPage schema to your top five pages. Rewrite the first paragraph under each H2 to be a direct 60-word answer to a specific question. That single move touches every assistant.
If you have one day:
Everything above, plus verify your site in Bing Webmaster Tools and Brave Search Webmaster Tools. Submit sitemaps to both. Allow OAI-SearchBot and Claude-User in your robots.txt.
If you have one week:
Everything above, plus pick five of your highest-value buyer questions, publish a dedicated page answering each one in answer-first format, with named author, publication date, external citations, 15+ entities per page and full schema. These are the pages the next 90 days of AI citations come from.
What this does not cover
This guide is tactics, not strategy. It does not cover which buyer questions to target, how to track whether the tactics are working, or how to price the engagement if you are hiring someone to do it. For those, see our resources: the AI SEO Glossary for the terminology and the 50 AI SEO Questions template for a starter prompt set.
For full service, see our AI SEO page.