// Answer
How do I rank in ChatGPT?
You don’t — not in the SEO sense. ChatGPT doesn’t rank pages; it retrieves, cites, and echoes. Four levers decide whether it names you: schema, llms.txt, entity completeness, and citable content.
// Premise
ChatGPT doesn’t have a ranking algorithm.
The question conflates two things. There is no ChatGPT SERP, no PageRank-equivalent, no public scoring function that orders results. What there is is a generation process — and that process draws on three different sources of facts about your brand. Live retrieval through the browsing tool, which is Bing-backed and queries the open web at request time. In-context retrieval from sources the model pulled into a specific response. And statistical recall from the training corpus, frozen at whatever cutoff the active model version uses. You “rank” in ChatGPT the way a quote ranks in a journalist’s article: you’re named because something about you made the writer reach for you. The job is to be reachable.
// Mechanics
What ChatGPT cites and why.
When ChatGPT browses, it issues a query, ranks candidate URLs (Bing’s ranker, mostly), fetches a handful, extracts spans, and feeds them into the prompt. Two things decide whether your page makes that shortlist: how the search engine ranks it, and how legible the page is once fetched. The first is regular technical SEO — crawlability, freshness, link graph, query-intent match. The second is new: schema, headings, claim density, semantic structure. A page that’s ranked #4 on Bing but parses cleanly often beats a page ranked #1 that hides its content inside a JavaScript framework. When ChatGPT isn’t browsing, it falls back to training-data echo: your brand is named because it appeared often enough, in clearly enough described contexts, that the model learned the association. That part you influence by being everywhere — consistently, with the same name and the same category claim.
// Levers
The four levers that actually move the needle.
- Structured data. JSON-LD for Organization, WebSite, Article, FAQPage, HowTo, Product, Offer, BreadcrumbList, and DefinedTerm. The engine doesn’t guess what your page is about when you’ve told it.
- An llms.txt manifest. A flat, parseable list of your canonical content. Ingestion pipelines prefer it because it’s cheaper to crawl than your sitemap and richer than your robots.txt.
- Entity completeness. Your brand should resolve as one entity, not three. Same legal name on Crunchbase, Wikidata, LinkedIn, and your site. Same one-sentence category claim on your About page, your G2 listing, and your Wikipedia article.
- Citable content. Short, specific, claim-dense sections an LLM can lift into an answer without rewriting. “The five engines that matter are X, Y, Z, A, B” beats “In today’s rapidly evolving AI landscape…” every time.
// Mistakes
Common mistakes.
- Stuffing prompts into H1s. ChatGPT extracts entities and claims, not phrases. “Best CRM for startups 2026” in your H1 doesn’t make you the best CRM for startups in 2026.
- Ignoring schema. The single largest gap between brands that get cited and brands that don’t is JSON-LD coverage. It is unglamorous, mechanical work, and it dominates.
- No llms.txt. It costs an afternoon to ship and it’s the cheapest legible signal you can send.
- Optimizing for ChatGPT only. The brand that wins ChatGPT often loses Perplexity. Operate cross-engine or get blindsided.
// Measurement
How to know if it’s working.
Search Console can’t see this. You need a tracker that runs your category’s prompts against ChatGPT — with browsing on and off — logs whether your brand is named, and watches the deltas. Tracking AI citations covers the mechanics. Vizelo runs this hourly across ChatGPT and the five other engines that matter, then surfaces the prompts where your share of voice is moving up, down, or flat. Start free to see your current footprint.
FAQ
Common questions.
Is there a ChatGPT ranking algorithm I can optimize for?
No. ChatGPT decides what to say through a mix of training-data recall, retrieval (its browsing tool is Bing-backed), and the in-context sources it pulls into a single response. There is no public PageRank-equivalent. You optimize the inputs — schema, llms.txt, entity graph, citable content — not a ranking function.
How long until ChatGPT starts citing me after I make changes?
Browsing-tool retrieval reflects changes within days, sometimes hours. Training-data echo only updates with new model versions, so changes there can take months. Most of the lift you’ll measure in the first quarter comes from the browsing path.
Do I need a Wikipedia page to rank in ChatGPT?
No, but you need entity completeness — your brand needs to be findable and described consistently somewhere the engine trusts. Wikipedia and Wikidata help, but well-structured Crunchbase, LinkedIn, and G2 pages plus an Organization schema on your own site cover most of the gap.
What schema types matter most for ChatGPT visibility?
Organization and WebSite for the site itself; Article, FAQPage, and HowTo on content pages; Product and Offer on commercial pages; DefinedTerm on glossary entries; BreadcrumbList everywhere. The goal is to hand the engine pre-parsed facts so it doesn’t have to guess.
Can I rank in ChatGPT without writing new content?
Sometimes. Adding schema, fixing entity issues, and shipping an llms.txt can lift citations on existing content. But if your category’s winning answers are claim-dense and yours are vague, you’ll need to rewrite for citability.