How Do You Rank in ChatGPT and Perplexity? A GEO Playbook for B2B/SaaS

"Ranking" in ChatGPT and Perplexity is not the same game as ranking in Google. There is no position 1. There is a generated answer, and either you get quoted inside it or you don't. This guide is the plain-language version of how we approach it at Greeto: what AI answer engines actually reward, the myths worth ignoring, and a cheap way to check if any of it is working.

How do you rank in ChatGPT and Perplexity?

You rank by getting cited inside the generated answer. AI engines retrieve short passages, not whole pages, then stitch the best ones into a reply. So you win by writing clear, self-contained answer chunks, publishing original data worth quoting, adding schema so machines parse you cleanly, keeping pages fresh, and allowing AI crawlers to read your site. Citations, not positions, are the goal.

This is the discipline behind two terms you'll see a lot. Answer engine optimization is about being the source a direct answer is built from. Generative engine optimization is the same idea aimed at generative tools like ChatGPT, Perplexity, and Google's AI Overviews. Both replace the old goal — "be the top link" — with a new one: "be the quoted source." Everything below serves that one goal. For the B2B angle specifically, our AEO for B2B websites playbook goes deeper on intent and buyer questions.

AEO vs GEO vs SEO: what's the difference?

SEO earns ranked links. AEO earns the direct answer. GEO earns a citation inside AI-generated text. They share a foundation — crawlable, fast, well-structured content — but the unit of success differs: a position, an answer box, or a quoted passage. You don't pick one. You do classic SEO well, then add the answer-first and citation layers on top.

Here is how the three compare in practice:

The practical takeaway: SEO and AEO feed GEO. A page already winning a featured snippet is usually well-positioned to be quoted by an answer engine too, because both reward the same tight, extractable answer.

How do AI answer engines pick sources?

Answer engines retrieve at the passage level, not the page level. They embed your content, match chunks to the user's question, prefer sources that clearly cover the entities involved, and lean toward fresh, trustworthy pages. A 3,000-word page can lose to a single crisp paragraph elsewhere — because the engine quotes paragraphs, not URLs.

Three things matter most here. First, passages: each section must answer one question completely, with no pronouns pointing back to earlier sections, because the engine may lift it alone. Second, entities: name the product, company, category, and concepts plainly and consistently so the model knows what your page is "about" — vague writing gets skipped. Third, freshness: many AI features over-weight recent content, especially for fast-moving topics, so a visible, real update date helps. Retrieval also favors sources that are easy to parse, which is exactly where schema and clean HTML earn their keep. If you're moving off a slow or messy stack to fix this, our WordPress to Next.js migration SEO guide covers doing it without losing rankings.

Does llms.txt help you rank in ChatGPT? (The myth-bust)

Mostly no — and this is the most over-hyped GEO tactic of the moment. Google has publicly said it does not use llms.txt for Search or AI Overviews, and Google's John Mueller compared it to the long-dead keywords meta tag. Adoption by other AI crawlers is still limited and inconsistent (Perplexity is the notable exception). So adding llms.txt will not make ChatGPT or Google rank you. Anyone selling it as a ranking lever is selling hope.

Here's the honest picture of where the file does and doesn't earn its place:

So why do we still ship it on some sites? Because the real value is the agentic layer: tools like Claude Code and other coding agents can use a curated llms.txt to navigate your docs efficiently. That's a developer-experience win for technical products, not a search-ranking win. We dug into the evidence in does llms.txt help ChatGPT. Treat it as docs infrastructure, not an SEO trick — and never let it replace the work below.

How do you write passages that get quoted?

Lead every section with a 40–60 word direct answer that stands alone. The engine often quotes only that opening, so it must be complete, accurate, and free of "as mentioned above." Put the answer first, then the explanation, then the nuance — the inverted pyramid. Use the literal question as the heading so retrieval can match it.

A reliable structure for an answer-first chunk:

1. Heading = the real question. Use the exact phrasing a buyer types, not clever wordplay.
2. First sentence = the whole answer. Someone could quote it and be correct.
3. Next 2–4 sentences = the proof. Add the "why," a number, or a concrete example.
4. No orphan pronouns. Don't open with "this" or "it" referring to a previous section.
5. One idea per chunk. Keep each section 100–300 words so it's easy to lift cleanly.

This is the same instinct that wins featured snippets, which is why AEO and GEO reinforce each other. For B2B specifically, structure your highest-value pages — pricing logic, comparisons, "how X works" — as stacks of these chunks. Our take on building a B2B homepage that converts shows how answer-first writing also helps human buyers, not just machines.

What content gets cited most by AI engines?

Original data. When you publish a number, benchmark, survey result, or first-hand finding that exists nowhere else, you become the only possible source — so the engine has to cite you to use it. Princeton's GEO research (Aggarwal et al., "GEO: Generative Engine Optimization," KDD 2024) found that adding statistics, citations, and quotations was associated with content being surfaced more often in AI-generated answers.

The hierarchy we see, roughly highest to lowest citation pull:

You don't need a giant research budget. A small, honest dataset — "we measured X across our own projects and here's what we found" — beats a long post of recycled tips. One caution, and it's the Greeto rule: only publish numbers you actually have. Inventing a statistic to get cited is the fastest way to lose trust the moment someone checks.

Is schema markup needed to rank in AI search?

It's not strictly required, but it's the cheapest way to make your content machine-readable, so treat it as citation infrastructure. Schema markup (structured data in JSON-LD) labels what your page is — an article, an FAQ, a product, an organization — so engines parse entities and answers without guessing. Clean structure lowers the cost of quoting you.

The high-value types for B2B/SaaS are Article (with a real author and dates), FAQPage for genuine question-and-answer blocks, Organization to define your brand entity, and Product or SoftwareApplication for the thing you sell. Pair an FAQPage block with on-page answer-first text and you've built the same content twice — once for humans, once for machines. Two cautions: mark up only content that's actually visible on the page, and never fake FAQs just to add markup. We treat schema, sitemaps, and crawler rules as one infrastructure layer; the WordPress to Next.js migration SEO guide shows how we wire it on a modern stack.

How do you let AI crawlers read your site?

Explicitly. Many sites accidentally block the exact bots that feed answer engines, then wonder why they're never cited. If you want to be quoted, allow the relevant AI crawler user-agents in robots.txt — and confirm your pages render their content in HTML, not only after heavy JavaScript that bots may not execute.

A simple procedure:

1. List the bots that matter. For citations, that's agents like OAI-SearchBot and PerplexityBot (the search-and-cite crawlers), plus Google's standard crawler for AI Overviews.
2. Decide training vs. answering. You can allow search/answer bots while blocking pure training crawlers (like GPTBot) if you prefer — they're different user-agents with different jobs.
3. Allow, don't silently block. Make sure no blanket Disallow: / or firewall rule is shutting them out.
4. Check rendering. View the raw HTML; if your key passages only appear after client-side JS, server-render them so bots see the text.
5. Verify in logs. Watch server logs to confirm the bots are actually fetching pages.

This is a one-time setup that gates everything else — the best passage in the world can't be cited if the crawler can't read it.

Do third-party mentions help you rank in ChatGPT?

Yes. Answer engines synthesize across the web, so being named on sites you don't own — directories, review platforms, roundups, reputable articles, community threads — raises how often models associate your brand with a topic. A page on your own domain is one vote; consistent mentions elsewhere are many votes that reinforce your entity.

The practical move for B2B/SaaS: get listed accurately where buyers compare tools, earn mentions in genuinely useful third-party content, and keep your name, category, and description consistent everywhere so the model sees one coherent entity instead of a fuzzy one. You can't fake your way here, and you shouldn't try — but you can make sure that where you're already discussed, the facts are right and current. Off-site consistency plus a strong on-site entity is what makes a model confident enough to cite you by name.

How do you track AI citations cheaply (weekly)?

Ask the engines your buyers' real questions and log who gets quoted. You don't need a paid tool to start. Pick 15–25 high-intent questions, run them through ChatGPT, Perplexity, and Google AI Overviews on a fixed day each week, and record whether you're cited, who else is, and which passage they pulled. That trend line is your GEO scoreboard.

A repeatable weekly routine:

1. Build a question set. 15–25 prompts a buyer would actually type ("best tool for X," "how does Y work," "X vs Z").
2. Pick a fixed cadence. Same questions, same day weekly — consistency beats volume.
3. Log five fields per run. Date, engine, question, cited? (yes/no), and which competitor was cited.
4. Capture the quoted passage. Note the exact wording so you can see what phrasing wins.
5. Act on gaps. Where a competitor is cited and you're not, write or sharpen an answer-first chunk for that question, then re-check next week.

A plain spreadsheet is enough to see movement in a month. Track citation frequency and your "share of voice" against two or three competitors, and let that decide what to write next. This closes the loop: you ship answer-first passages and data, allow the crawlers, then measure which questions you own — and which you still need to win.

FAQ

Can I pay to rank in ChatGPT or Perplexity?

No. There is no ad slot or paid placement that buys you a citation in an organic AI answer. You earn it by being a clear, trustworthy, quotable source. Some engines run separate ad products, but those are labeled and distinct from the cited sources in the generated answer.

Does llms.txt help me rank in Google or ChatGPT?

No. Google has stated it does not use llms.txt for Search or AI Overviews, and John Mueller likened it to the obsolete keywords meta tag. Its genuine value is for agentic and IDE tools — giving coding assistants a clean map of your docs — and some experimental support in Perplexity. Treat it as documentation infrastructure, not a ranking lever.

How long does GEO take to show results?

It varies, and anyone promising a fixed timeline is guessing. In our experience, fixing crawler access and adding schema can register within weeks, while building citations through original data and third-party mentions is a months-long effort. Track weekly so you see the trend instead of waiting for a single big moment.

Is GEO different from regular SEO?

They overlap but aim at different targets. SEO ranks pages; GEO earns citations inside AI-generated answers. The foundation is shared — crawlable, fast, well-structured, trustworthy content — but GEO adds answer-first passages, original data, schema, and citation tracking on top of solid SEO. Do both; one feeds the other.

What's the single highest-impact GEO move for a B2B site?

Publishing original data nobody else has. When you own a number, a benchmark, or a first-hand finding, an engine has to cite you to use it. Combine that with answer-first passages and allowed crawlers and you've covered the three things that move citations most. Just make sure every number is real.

Do I need schema markup to be cited by AI?

Not strictly, but it's the cheapest way to help. Schema markup makes your content machine-readable so engines parse your entities and answers without guessing, which lowers the effort to quote you. Use Article, FAQPage, Organization, and Product types — and only mark up content that's genuinely on the page.

How do I know if my GEO work is actually working?

Run a fixed set of buyer questions through ChatGPT, Perplexity, and AI Overviews on the same day each week and log whether you're cited. Watching citation frequency and share of voice over a month tells you far more than any one-off check. Where a competitor wins, write a sharper answer-first chunk and re-test.