How to rank in ChatGPT (and what 'rank' actually means)

The short answer

Ranking in ChatGPT is a different problem from ranking in Google. ChatGPT search uses Bing's index for retrieval, then synthesizes an answer across multiple sources. There is no position number. The win condition is being one of the two or three sources cited in the answer paragraph.

To get cited, three things have to be true. Your page is in Bing's index and reasonably ranked for the query. The page contains specific extractable facts that match the question. Your business has authority signals, including brand mentions across independent contexts on the open web.

Most of the work is on-page content discipline plus a slow accumulation of brand mentions. The vendors who promise to "rank you in ChatGPT" through a tool feed are selling a fiction. The actual mechanism is closer to traditional SEO with a different rubric on top.

How ChatGPT search actually works

As of 2026, ChatGPT search has two retrieval layers. The first is Bing's web index, which provides the underlying pool of pages the model can read. The second is a custom retrieval layer that scores and selects the pages most relevant to the query, then feeds three to ten of them as context into the model that composes the answer.

The model produces a synthesized response, citing the two or three pages whose content most directly informed the answer. Citations are visible: each cited source shows up as a numbered link below or inline with the answer, depending on the surface (web ChatGPT, app, or API).

For ranking, this means three filters apply.

Indexability. The page has to be in Bing's index. If your robots.txt blocks Bingbot, or your site is technically broken, you start out invisible. Verify your site is crawled by Bing through Bing Webmaster Tools.

Retrieval relevance. The page has to score high enough on the custom retrieval layer to make it into the context window of the answering model. This layer weights signals similar to traditional ranking (authority, relevance, freshness) plus signals specific to AI answer synthesis (factual density, structured data, citation-friendliness).

Citation selection. Once in the context, the model picks two or three sources to cite based on which pages contributed most to the synthesized answer. Pages whose content the model paraphrased or quoted directly tend to win the citation.

Signals that drive citations

Empirically, across hundreds of test queries on ChatGPT search throughout 2025 and 2026, pages that consistently win citations share specific qualities.

Concrete numbers and entities. Pages that say "foundation repair in the Pacific Northwest typically costs $4,000 to $18,000 depending on soil conditions" beat pages that say "foundation repair pricing varies by project." The first version gives the model a fact to extract and attribute. The second gives nothing.

Direct answers to common questions. Pages that include question-format H2s with tight answer paragraphs underneath get cited more often than pages that bury the answer in prose. The structure mirrors how the model breaks the question down internally.

Original data. Survey results, project counts, average timelines, geographic patterns, anything you have that nobody else has. AI engines preferentially cite original data because the alternative is paraphrasing the same source the user could read directly.

Authority signals on the page. A named author with a bio. Credentials that match the topic. Real address, real phone, professional registration. Schema markup that confirms the business is real.

Brand mentions across the open web. The model's training and continuous updates absorb mentions of your business in newspapers, trade publications, podcasts, expert roundups, and industry directories. A foundation repair contractor mentioned in five regional papers, two trade publications, and a Reddit thread is more likely to be cited than one mentioned only on its own site.

llms.txt. A new file at your domain root that gives AI crawlers a curated map of your most important content. Adoption is still early, which means first movers get a small but real advantage in retrieval relevance.

Step-by-step optimization

For a service business that wants to start getting cited in ChatGPT inside 90 days, the work is sequenced.

Step 1, weeks 1 to 2. Audit indexability in Bing Webmaster Tools. Verify your site is crawled. Submit your sitemap. Fix any crawl errors.

Step 2, weeks 1 to 3. Audit your top ten pages for factual density. Each page should include at least five concrete facts (numbers, dates, dollar figures, named entities) relevant to the topic. Add what is missing. Cut what is filler.

Step 3, weeks 2 to 4. Restructure your top ten pages so each H2 is a question and each H2 is followed by a 40 to 60 word direct answer paragraph. The rest of the section continues as before.

Step 4, weeks 2 to 4. Add full schema markup. LocalBusiness, Service, FAQPage, Article, Person on author bios. Validate using Schema.org's tools.

Step 5, weeks 4 to 8. Publish llms.txt at your domain root. Include your hub pages, your most authoritative service pages, and your top guides.

Step 6, weeks 4 to 12. Pursue brand mentions. Pitch one regional publication, one trade association, and one industry directory each month. Aim for five new independent mentions in the first 90 days.

Step 7, ongoing. Track citations. Run your top 20 buyer queries through ChatGPT search every two weeks. Note which queries cite your business. Trace what changed when citation counts move.

What to test, what to track

Citation tracking in 2026 has improved but is still rough. Three options exist.

Manual testing. Run a list of 20 to 50 buyer queries through ChatGPT search every two weeks. Record which queries cite your business and which compete with you. Free and reliable but labor-intensive.

Profound. A SaaS that tracks LLM citations across ChatGPT, Perplexity, Claude, and Google AI Overviews. Pricing starts around $300 a month for small businesses. Useful for at-scale tracking and competitive monitoring.

Otterly. Similar tracking tool, lighter pricing for small budgets, narrower coverage. Useful for getting started.

For a service business with a focused query list of 20 to 50 terms, the manual approach often works fine and is the least expensive way to learn what is moving citations on your specific buyer questions.

Common failure modes

Three patterns produce flat results despite consistent work.

Generic content. Pages full of vague positioning language and no specific facts produce no citations because the model has nothing to extract. Diagnostic: read your service page out loud. If you can replace your business name with a competitor's and the page still works, the page is too generic.

No off-page signals. A site with great content but zero independent mentions on the open web has a hard ceiling. The model's authority weighting reads it as small or new and prioritizes other sources. Diagnostic: search your business name in quotes. If you only appear on your own site, you have an off-page problem.

Schema mistakes. Schema that validates technically but misrepresents the business (wrong type, wrong entity relationships) confuses retrieval. Diagnostic: paste your schema into Google's Rich Results Test and Schema.org's validator. Confirm the entity graph reflects what you actually do.

Most stuck programs have one of these three problems. The fixes are concrete and usually take 60 to 120 days to compound into citation gains.

What this is worth

For a service business doing $500k to $5M in annual revenue from leads, the math on ChatGPT citations works fast. A typical citation pattern is one to two cited mentions a month per ten queries you optimize for. At buyer-research queries, citation share roughly translates to top-of-funnel awareness, which compounds into call volume over 6 to 12 months.

The cost is your team's time, plus optionally a $50 to $500 a month tracking tool. The payback, in additional booked work, usually shows up between months 4 and 8 for a service business with average ticket above $5,000.

The window is open now because most service businesses are not running this work yet. The window will close in the same way SEO closed between 2003 and 2010, with first movers locking in citation patterns that take years to displace.