How to rank in ChatGPT starts with being in the answer at all

To rank in ChatGPT you have to be the brand it names when a buyer asks it what to buy. Not cited in a footnote. Named in the sentence. That is a different job from ranking on Google, and most B2B brands are not doing it yet, because they are still optimizing for a search results page their buyers increasingly skip.

Here is the shift in one line. Buyers used to Google a category, click a few vendor sites, check G2, and ask a peer. Now they open ChatGPT, Perplexity, or Gemini and ask “best [category] for [use case],” and they arrive at your site, if they arrive at all, already holding a shortlist the model built for them. Roughly 47% of B2B buyers now start vendor research in an AI tool before visiting any website. If you are not in that answer, you are not on the shortlist, and your Google ranking never gets a vote.

I run this across six B2B SaaS accounts, in cybersecurity, HR tech, legal, observability, creator tools, and marketing tech. This is the playbook I actually use, the four levers that move it, the three situations you will recognize, and how to measure whether it is working.

That second number is the point. This is not a traffic-recovery exercise. LLM-referred visitors show up further along, already sold on the shortlist, and they convert several times better than a cold organic click. On one creator-tools account I run, LLM sessions went from 11,639 in April to 80,363 by mid-June, and they arrived alongside record subscriptions. The click you lost to a zero-click answer was often the low-intent one. The visit you win from ChatGPT is the buyer.

Ranking in ChatGPT means getting named, not just cited

Ranking in ChatGPT is not the same as being one of the little citation links under the answer. A citation is a source the model read. A recommendation is the brand it actually names. Buyers act on the name, skim past the links, and a model will happily cite your own “best [category]” page while recommending a competitor inside the answer.

So the scoreboard is share of the named recommendation, not citation count. I have watched an account be the single most-cited domain in its category and still lose the recommendation to rivals, which is the whole argument in AI citations are a vanity metric. Everything below is aimed at the name, not the footnote. And it is broader than Google: the same work moves ChatGPT, Perplexity, Gemini, and Claude, which is why this is a ChatGPT playbook and not just an AI Overviews playbook.

Four levers move it. Coverage density, technical GEO, off-site presence, and PR. Pull one and little happens. Pull all four and you become the default answer.

Lever 1: coverage density and topical authority

LLMs favor brands with real depth on the topics buyers care about, not a homepage and a prayer. If every question a buyer might ask has a substantive page behind it, the model has something of yours to pull for each one. If it does not, it pulls someone else’s.

Coverage density means four kinds of pages, built out deliberately:

  • Core solution pages for every capability, use case, and industry you serve.
  • Prompt-match content for the narrow, mid-to-low-funnel questions buyers actually type: “X pricing for Y,” “how does X integrate with Z,” “X vs Y for [use case].”
  • Comparison and alternatives pages. Ranking in “[X] vs [Y]” and “best [category]” content is a primary path to being pulled into an AI answer.
  • Thought leadership with a point of view and proprietary data. LLMs quote a distinct take; they summarize and discard commodity information they can find on fifty other pages.

That last one is where Information Gain does the heavy lifting. Google’s ranking signal of 2026 measures how much genuinely new knowledge your page adds relative to what already ranks, and the March and May core updates re-weighted it hard. I watched this play out on a legal SaaS account I run: the blog playbook that won in 2024 and 2025 (cover the topic, publish a long ten-item list) is the exact format that got demoted. Length and keyword coverage are table stakes now; a defensible argument plus first-party proof is what survives and gets cited.

This works, and I can show the receipts. At Directive, we ran exactly this coverage-density playbook on our own site: built out our core service pages, shipped 150 new articles in a single quarter, added a set of “best-of” listicles, and earned third-party citations. The result over roughly a year:

Note which number is the win: 24% share of AI brand mentions, roughly twice the next competitor. That is being named, at scale, which is the entire game.

Lever 2: technical GEO, so AI can actually read you

Coverage means nothing if the crawlers cannot absorb it. GEO starts with content, but technical foundations decide whether an AI can ingest it, and this is the lever most teams skip. Four checks:

  • Crawl accessibility. Confirm you are not blocking GPTBot, PerplexityBot, or ClaudeBot in robots.txt, and use Search Console to stop wasting crawl budget on useless pages. On a cybersecurity account I run, clearing tracking-parameter and feed cruft freed crawl for the pages that mattered and helped resolve roughly 280 “crawled, currently not indexed” URLs.
  • Schema markup, stacked. Structured data helps LLMs understand what a page is and how it fits together. Stack multiple relevant types per page rather than settling for one. On the same cybersecurity account, a schema-only rollout drove a +45% month-over-month click gain with no new content, the cleanest single-variable proof I have.
  • Server-side rendering. JavaScript-rendered content is invisible to many AI crawlers. If your content only appears after the JS runs, render it server-side or pre-render it, or the model never sees it.
  • Absorbable page structure. Clean H1/H2/H3 hierarchy, FAQs, summary paragraphs, definition-first writing, and TL;DR boxes let a model extract and quote you cleanly. This is the on-page half I break down step by step in how to optimize content for LLMs.

Lever 3: off-site presence, because you are 3% of your own story

Here is the lever that reorganizes the whole plan. Across the six accounts I track, a brand’s own website is only about 3% of the sources AI cites when it answers questions about that brand’s category. The other 97% is third-party, and a chunk of it is your direct competitors. I published the full breakdown in the generative engine optimization statistics, and the takeaway is blunt: you cannot win the answer from your own homepage, because the model barely reads it.

So you have to show up where the model actually looks. In practice that is a short, stable list:

  • Reddit. LLMs lean on it heavily, and Google ranks Reddit threads aggressively for vendor-evaluation queries. Try searching “best MDR” or a category-plus-review query and watch Reddit sit in the top three. On my cybersecurity account, competitors get named constantly in those threads and the client much less, and, in the account’s own words, that thin third-party presence is directly capping its LLM citations. Participate 90/10, value first, and own the “best [category]” threads honestly.
  • YouTube how-tos and demos, with real transcripts the model can read.
  • LinkedIn articles and posts from real people. Repurposing owned BOFU content into LinkedIn articles from several employee profiles is a low-lift LLM-visibility test I have run directly.
  • Entity corroboration. A clean Wikipedia and Wikidata record, consistent G2, Capterra, and Trustpilot reviews, and accurate directory facts (Crunchbase) give AI a stable, trusted picture of who you are.

Off-site moves slowly and feels un-trackable, so it gets neglected. It does not have to. Here is a monitoring stack you can stand up this week:

  1. Run site:reddit.com searches for your brand, products, and exec names.
  2. Append .rss to any subreddit, search, or profile URL to pull a free feed.
  3. Point F5Bot at your keywords for free email alerts across Reddit and Hacker News.
  4. Pipe every alert into one Slack channel.
  5. Assign one person to triage it daily, because alerts only work if someone reads them.

Lever 4: PR and third-party, the citations models trust most

A press release on your own blog does nothing on its own. What moves an LLM is off-domain pickup from sources it already trusts, and third-party citations carry far more weight than raw links. A mention in a respected industry publication outweighs a pile of low-quality backlinks. Three moves:

  • Lead with digital PR, timed to launches. Product rollouts and data drops are your strongest hooks for editorial coverage on the domains LLMs ingest.
  • Run original research on a cadence. A quarterly pulse, a semi-annual deep dive, an annual benchmark. Slice one study by industry and company size for many citable cuts. Proprietary data others cannot replicate wins twice: it lifts rankings and makes you the primary source models cite. That is the first-party data moat in practice.
  • Target third-party listicles like a media buy. Audit the SERP for your category to see what ranks, Gartner, G2, industry roundups, Reddit. Find the path into each, whether that is paid placement, editorial outreach, or review velocity. Then run the ROI math and pay only where the numbers work. Your rivals will rarely add you to their own lists, so aim at the neutral, authoritative ones buyers and models actually read.

Three situations you might recognize

The levers are the same everywhere. How you weight them depends on the problem. Three come up constantly.

Situation 1: You are moving upmarket, but ChatGPT still sees an SMB tool

The model holds a calcified view of you, built from years of SMB-flavored content, review snippets, and SMB pricing pages. So when a buyer asks for “best [category] for enterprise,” you are absent, even when your enterprise product is genuinely strong. The model is not being unfair; it is reflecting the picture the web has painted of you, and that picture is years behind your go-to-market.

Fixing it is bigger than a messaging refresh, and this is where teams underestimate the work. You have to rewrite essentially everything across your own site first. Not just the homepage hero, every cornerstone page, solution page, and comparison page has to be re-cut in enterprise vocabulary: SSO, SOC 2, RBAC, procurement, security review, implementation, and the buyer titles that come with a six-figure deal. Then you build the enterprise proof the model looks for, named enterprise logos, compliance and security content, case studies with enterprise outcomes, customer councils, and G2 reviews written in enterprise language instead of SMB language.

But rewriting your own site is only half of it, and it is the smaller half. Remember lever 3: you are roughly 3% of your own story. The calcified view lives out on third-party sites and social, so you have to go change the conversation where it actually happens. That means reaching back out to every place you are already published or mentioned and getting the framing updated to match your new position. It means seeding the enterprise narrative on Reddit, LinkedIn, and YouTube through real people and organic social, so the model sees enterprise buyers discussing you in enterprise contexts. And it means digital PR aimed squarely at enterprise-credible outlets, because a placement in a publication your enterprise buyer trusts is what re-teaches the model who you are. That off-site work is what actually gets pulled into the answer. Give the whole program 90 to 120 days before the enterprise prompts start to move, because you are not editing pages, you are re-training the web’s memory of you.

Situation 2: You and a competitor are even on SEO, but they get named 2 to 3x more

This is the most common and most frustrating one, because on paper you are matched, similar domain authority, comparable rankings, similar content, and yet the model keeps naming them and skipping you. The instinct is to go write more pages, but that is guessing. The first job is to see what specifically is getting pulled in when the model recommends them. Run the actual prompts, read the citations underneath, and the gap will be one of three things:

  1. It is a third-party gap. Very often the source the model is leaning on is not even one of their pages. It is a third-party site: a Reddit thread, a G2 grid, an industry roundup, a review page where they are named and you are not. If that is the gap, the fix is not on your site at all. You have to go get onto more of those third-party sites and get named in the same places they are. Reverse-engineer their citation footprint and earn the same placements.
  2. It is a structure gap. Sometimes it is their own page, but the difference is structure, not substance. Their page is simply easier for the model to extract, cleaner hierarchy, a direct definition up top, a comparison table, an FAQ, a crisp summary, so it gets quoted and yours does not. If that is the gap, you match them: restructure your equivalent page until it is at least as extractable as theirs. You are not trying to out-write them, you are trying to be at least as easy to pull from.
  3. It is a proof gap. Sometimes they have proprietary research, a benchmark, a data study, a number nobody else can cite, and the model rewards it because it is genuinely new information. If that is what separates you, there is no shortcut, you have to get your own original research started, because that is the asset that earns citations you cannot buy.

One useful tell across all three: if your citations are rising faster than your named mentions, your content is sitting too high in the funnel, so balance it with brand-led, bottom-funnel pages.

Situation 3: Organic is down 20 to 40% and you want it back

Do not chase the lost clicks one for one; a lot of them were low-intent TOFU that AI Overviews now answer on the page. The clicks are gone, and most of them were never worth much.

Here is the part I will say twice, because it is the whole problem. If you are a CMO or a head of marketing holding your team to clicks as the metric, you will never see pipeline. If you are holding yourself to clicks, you will never see pipeline. When clicks are the scoreboard, your team writes to the scoreboard: they pump out a pile of high-volume informational garbage because it draws traffic, and almost none of it converts. Some of the highest-volume topics in your category are some of the lowest-converting, because they are purely informational and carry zero buyer intent. You can win that traffic and still lose the quarter. Clicks measure how many people showed up; they say nothing about whether any of them were buyers.

So reframe around qualified pipeline from any AI-influenced source, and the picture flips. This is the pattern I see every month: on an HR-tech account, MQLs held steady through real traffic declines because the high-intent buyers were still finding the site and still converting, and its bottom-of-funnel clicks recovered while the top of funnel bled, exactly the BOFU-bucks-the-trend split. On a cybersecurity account, the new content folder kept growing straight through the volatile May core update, its top-three keywords more than doubling, and it became the engine of the site’s recovery. Inventory your high-intent prompts, build landing experiences for someone an LLM already warmed up, and shift spend from generic TOFU toward comparison, alternatives, and proof.

How to measure ChatGPT rankings

You cannot improve what you cannot see, and “are we in ChatGPT” has a real answer now. Four sources give you the full picture:

  • Brand mention and share of voice. How often you appear in LLM answers versus your top three competitors, tracked week over week. This is the scoreboard.
  • Referral analytics and CVR. Custom channel groupings in GA4 for ChatGPT, Perplexity, Gemini, and Claude referrers, tied to pipeline. This is where you prove the 2-to-4x conversion story and connect it to revenue, the same pipeline-over-sessions measurement I use everywhere.
  • Prompt-level citation tracking. Pick at least 30 priority prompts a real buyer would ask and track which brands appear, how often, and in what context, with trend data. This is the part you will not do by hand, and it is worth saying plainly: you are going to need a purpose-built third-party tool. Scrunch and Profound are the two I reach for most; they were built to watch LLM answers at the prompt level and show you share of voice over time. You can approximate some of this in Ahrefs, and it is shipping AI-visibility features fast, but this is a young, quickly moving category and the dedicated tools are ahead of it for now.
  • Crawler log analysis. Confirm GPTBot, PerplexityBot, and ClaudeBot are actually reaching your content. Crawl gaps are visibility gaps.

This whole measurement layer is getting built out in front of us. More of it becomes trackable every month, and I expect that to accelerate. My guess, and it is a guess, is that as the ad platforms and model providers race to monetize AI search, they will end up exposing far more about who is searching and what they are asking, the same way paid search eventually surfaced query and audience data. Privacy pressure pushes the other way, so I am not going to pretend I know exactly how it shakes out. What I do know is that the tooling is improving fast, and the teams that start measuring now are the ones who will be able to act on it when the data gets rich.

Attribution here is imperfect, and that stops a lot of teams from starting. Do not let it. The referral trend, the CVR gap, and the share-of-voice line are all measurable today, and the direction is undeniable. Start with measurement, make it real for your team with data they can see, and the budget conversation gets easy. If you want the mechanism behind all of this, it is the Great Decoupling: impressions and rankings hold while clicks fall, and the value moves to the answer.

FAQ

How does ChatGPT decide what to rank and recommend?

It leans on what already ranks in classic search, then favors sources it can cleanly extract and that corroborate each other across the web. That means your Google rank is the entry fee, your page structure decides whether you get quoted, and your off-site presence, reviews, Reddit, third-party roundups, decides whether you get named as the recommendation rather than just cited. No single factor wins it; the brand that shows up across all of them becomes the default answer.

How do I get my brand named in ChatGPT answers?

Be present in the third-party sources the model trusts, not just on your own site. Across the accounts I track, a brand’s own domain is only about 3% of the sources AI cites about its category, so getting named takes coverage density on-site plus Reddit, YouTube, LinkedIn, reviews, and earned media off-site. The name in the answer usually comes from somewhere other than your homepage.

Third-party citations matter more than raw links. A mention in a respected industry publication carries far more weight with an LLM than a stack of low-quality backlinks. Think in terms of authoritative, in-context mentions across the web, not link volume.

Is publishing more content the way to rank in ChatGPT?

Depth beats breadth. One definitive page with a point of view and proprietary data outperforms five thin pages that synthesize what already ranks, and the recent core updates actively demote the thin version. Cover the topics your buyers care about substantively; do not pad a content calendar.

How is ranking in ChatGPT different from ranking in Google’s AI Overviews?

The tactics overlap heavily, but the surface is broader. AI Overviews are Google-specific; ChatGPT, Perplexity, Gemini, and Claude each assemble answers from the open web and their own sources. The same four levers move all of them, which is why the goal is being the named answer across engines, not chasing one product’s citation box.

Can you pay to rank in ChatGPT, or is it free?

There is no ad slot or paid bid for the brand ChatGPT names, so in that sense ranking is “free”: it is earned, not bought. What it costs is the work behind the four levers, content, technical GEO, off-site presence, and PR, which is effort and time rather than a media budget. You cannot pay OpenAI to be recommended the way you pay Google for an ad, and that is exactly why being the named answer is durable once you earn it. Even as AI platforms begin testing ad formats, the organic recommendation stays earned, not purchased.