How to Rank in ChatGPT: A Step-by-Step Guide

ChatGPT has no page of ten blue links. When a buyer asks "best web-to-print software for a print shop", it writes an answer that names a few companies, often as a list, and — when it searched the web — links the pages it used. Ranking therefore means three measurable things: being named at all (mention rate), where in the list you are named (position), and whether your own page is one of the links (citation rate).

All three vary from run to run. The same question asked twice can name different companies, so a single screenshot proves little. Measure the rate over many runs, with an interval — a method we publish on our methodology page. Our own ChatGPT numbers for the Netbase JSC sites are on the ChatGPT SEO overview.

Access is necessary, and it is not enough. On 2026-11-09 we ran the ChatGPT Bot Access Test on 7 Netbase-owned websites (all part of Netbase JSC, the company behind Leadee) and on leadee.ai: all 8 allow GPTBot, OAI-SearchBot and ChatGPT-User by name, and none blocked the live fetches at the firewall.

Yet in Leadee's 2026-10-25 tracker round, OpenAI's API with web search named a Netbase brand in only 3 of 25 runs (12%, 95% interval 4-30%) — all three CMSmart, at an average position of 3.7 in the answer's list. Netbase JSC's own site, whose technical audit scored grade A, was named in 0 of 7. The OpenAI sample is small because the account ran out of credit mid-round; Perplexity and Google AI Mode, measured in full, showed the same pattern: only CMSmart and Printcart (both Netbase-owned) were named in answers to unbranded questions. The gap was off-site evidence and quotable content, not crawlability — which is why steps 5 and 6 below carry most of the work.

ChatGPT search finds pages through OpenAI's OAI-SearchBot. If your robots.txt blocks it — by name or through a catch-all User-agent: * / Disallow: / group — your pages cannot appear in ChatGPT search answers. GPTBot is a different bot, used for training, and blocking it does not remove you from search.

Check robots.txt and your firewall with the tool above, and read what each OpenAI bot does before you change a rule. The exact robots.txt lines and the CDN switch that most often blocks AI bots are in AI crawlers and robots.txt.

If your company facts only appear after JavaScript runs in a browser, many crawlers see an almost empty page. Put what you make or do, who you serve, where you operate, certifications, typical lead times and price ranges into the HTML the server sends on the first request. Product facts locked in a PDF catalogue or an image are invisible to a crawler that reads text.

Our guide to JavaScript rendering for AI crawlers shows how to check what a crawler actually receives.

Write down eight to twenty questions a buyer who does not know you yet would ask: "best [category] for [type of company]", "who makes [product] with [certification] in [country]", "[market player] alternatives". Questions that already contain your brand name are worth tracking, but they are not where new buyers come from.

The free ChatGPT Prompt Finder suggests questions for your industry. Keep the list stable once you start measuring, so rounds can be compared.

Run every question several times and record whether you are named, at what position, and which pages are cited. Do it through an official API rather than by copying answers from the ChatGPT app, which adds your account's memory and settings. Leadee's tracker runs the questions through OpenAI's Responses API with web search and reports each rate with a Wilson interval.

The baseline also shows who ChatGPT names instead of you and which pages it cites — the input for steps 5 and 6.

For each important question, make one page whose passages answer it on their own: the answer in the first sentence, the specific number you can document, who says so, and a named source. Spec tables in real HTML help too: a model extracts material, tolerance and lead time from a table more reliably than from a paragraph saying "a range of materials for most applications".

When we scored 766 passages on 8 Netbase-owned pages (part of Netbase JSC), 92% opened with a direct statement but only 5.5% carried a number and 2.5% named a source. The full method, with a before-and-after rewrite, is in how to write pages ChatGPT can quote.

For many buyer questions ChatGPT builds its answer from directories, review sites, app stores and market players' own pages, not from your site. In our data for the Netbase-owned sites, the most-cited domain for the outsourcing-services questions was clutch.co, ahead of Netbase JSC's own domain. An accurate profile and real client reviews on the two or three lists that are cited for your questions is often the fastest gain.

Add an honest comparison page for "[market player] alternatives" and "[A] vs [B]" questions, and keep one company name and one main domain everywhere. Step-by-step: how to get cited when ChatGPT prefers other websites.

Run the same questions the same number of times about four weeks after the changes go live, and compare rates with their intervals. Robots.txt and server rendering can take effect within days; directory profiles and new reviews follow each platform's own update cycle, often weeks. If a rate has not moved outside its interval, you do not yet know whether anything changed — run another round before drawing conclusions.

There is no prompt-injection trick, no hidden text aimed at a model instead of a human reader, and no way to buy a place in ChatGPT's organic answers. OpenAI does document ads in ChatGPT, but ads are a separate product from the sources an answer cites. Anything that promises a ChatGPT ranking for certain is describing something the technology does not offer.

Most of this guide is the same fundamentals good SEO has asked for years, applied to a different reader: be readable, be specific, be present where buyers and engines already look. For how these steps feed the retrieve-then-write process inside an AI engine, see how AI chooses which suppliers to recommend.