How to rank in ChatGPT in 2026: a step-by-step guide

This guide is for the B2B marketing lead, founder or SEO owner who has been asked a blunt question by the CEO: why does ChatGPT recommend our competitor and not us? It is narrower than our guide to AI search visibility, which covers Google AI Overviews, Perplexity, Claude and Copilot side by side. Here the focus is one assistant and one sequence of work.

By the end you should be able to pick the prompts worth competing for, check that OpenAI’s crawler can read your site, decide whether a given prompt needs a new page or a mention on someone else’s page, and report results in a way a finance team will accept.

It is not for brands whose buyers never research before buying, and it will not help a site that Google and Bing cannot index. Fix that first.

Summary

• “Ranking” in ChatGPT means two different things: being named in the answer (a mention) and being linked as a source (a citation). Track both.
• For questions that need current information, ChatGPT rewrites the prompt into one or more search queries and sends them to search partners. Those rewritten queries are what you compete for.
• Blocking OAI-SearchBot removes a site from ChatGPT search answers. Content that only appears after JavaScript runs is invisible to OpenAI’s crawlers.
• For “best X” and “X alternatives” prompts, comparison and list pages tend to be cited heavily, and many cited sources sit on other people’s domains.
• Answers vary from run to run, so a prompt needs weeks of repeated sampling before you can call a trend.
• Ads in ChatGPT are labeled and, per OpenAI, do not change the answers. There is no paid route into the recommendation itself.
• Clicks from ChatGPT undercount its effect. Measure mentions, citations, branded demand and pipeline together.

1. What “ranking in ChatGPT” actually means

ChatGPT does not show a list of ten blue links. It writes an answer, names a few brands and, when it searched the web, attaches citations to the sources it used. So “ranking” has two parts:

• Mention: your brand is named in the answer.
• Citation: one of your URLs is linked as a source.

The two do not always travel together. A page can be cited for a fact while the answer recommends three competitors. Practitioners call that a ghost citation, and it is common for brands whose blog is useful but whose product is not in the shortlists the model reads.

The audience is large enough to justify the work. OpenAI said in February 2026 that ChatGPT had reached 900 million weekly active users, up from the 800 million it reported in October 2025.¹1 Source: TechCrunch, “ChatGPT reaches 900M weekly active users”, accessed Oct 2026. Published February 27, 2026, reporting OpenAI’s announcement.

Model memory and live search

ChatGPT can answer from what it learned in training, or search the web first. OpenAI’s help center says it may search automatically “when your question would benefit from current information.”²2 Source: OpenAI Help Center, “Searching the web with ChatGPT”, accessed Oct 2026. Covers automatic search, query rewriting, search providers and site eligibility. Buying questions, such as “best CRM for a 20-person sales team”, usually qualify, because the user wants current products and prices.

That split matters for planning. Training data changes slowly and you cannot edit it. Live search reads whatever is on the web this week, so a new page can be cited within weeks, provided the crawler can reach it.

2. How ChatGPT picks the sources it cites

OpenAI has never released a formula. The closest thing is one line in its help center: that ChatGPT “ranks search results using multiple factors intended to help users find relevant, reliable information,” and that placement is not guaranteed.² What it does document is the mechanism around ranking, and that is enough to plan from.

Query rewriting, or “fan-out”

ChatGPT rarely searches the prompt word for word. Per OpenAI, when it uses search partners, ChatGPT “typically rewrites your query into one or more targeted queries” and may send further, more specific queries after reviewing the first results. OpenAI’s own example: a question about cancer drugs that target CCR8 might first become “CCR8 immunotherapy drug development 2025.” A prompt for good restaurants nearby can become “top restaurants San Francisco,” using approximate location from the IP address.²

The SEO trade calls this query fan-out. In practice it means one buyer prompt produces several searches, and the answer is assembled from the pages that win those searches.

Search partners and Bing

The same help page links to the privacy policies of Microsoft and Shopify as search providers ChatGPT may send queries to.² Microsoft’s Bing index is the one most B2B sites should care about. A page missing from Bing loses one of the routes ChatGPT uses to find it. Submit your sitemap in Bing Webmaster Tools and check its index coverage as carefully as you check Google’s. Bing’s AI Performance report also shows how often your pages are cited in Copilot and Bing’s AI answers, a useful proxy while ChatGPT offers no publisher console.³3 Source: Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview”, accessed Oct 2026. Covers Copilot and Bing AI answers, not ChatGPT.

3. Steps 1 and 2: choose the prompts, then watch them

Everything after this depends on picking the right questions. Spend more time here than feels comfortable.

Step 1: list the buyer prompts worth winning

Begin with questions buyers ask near a decision:

• “best [category] for [use case or company size]”
• “[competitor] alternatives”
• “[product A] vs [product B]”
• “is [product] worth it” and “[product] pricing”

These have less volume than educational questions, and far more of them end in a demo. Pull the wording from sales call recordings, support tickets, Reddit and Quora threads, review sites and Google’s People Also Ask boxes. Then rewrite your best-converting Google keywords as the full sentences people type into a chat box. Rank the list by the pipeline each prompt could influence, not by search volume, and start with the most valuable ones even when they are the most contested.

Skip brand-name prompts (“what is [your company]”) for now. They produce flattering numbers and tell you nothing about shortlists.

Step 2: sample every prompt for a month

Ask the same question twice and you may get two different shortlists. SparkToro and Gumshoe had volunteers run the same prompts nearly 3,000 times across ChatGPT, Claude and Google’s AI results in late 2025; in ChatGPT, getting an identical brand list twice in 100 runs was rarer than one chance in a hundred, by their estimate.⁴4 Source: SparkToro, “NEW Research: AIs are highly inconsistent when recommending brands or products”, accessed Oct 2026. Published January 28, 2026; 2,961 runs; not peer reviewed; co-author works at an AI tracking vendor. A screenshot of one answer is an anecdote.

Run each target prompt repeatedly, ideally daily, in ChatGPT and in the other assistants your buyers use. For each run, log:

• which brands were mentioned, and in what order
• which URLs were cited
• which search queries the model ran, where the interface shows them

Four weeks of this gives you a baseline and, more usefully, a list of the pages and domains that keep showing up. That list drives steps 4, 7 and 8. Manual sampling in a spreadsheet works for 20 to 30 prompts. Beyond that, a tracking tool or a managed service is cheaper than the staff time.

4. Step 3: make sure OpenAI can read your site

This is the cheapest step and the one most often skipped. If it fails, nothing else in this guide matters.

Allow OAI-SearchBot

Search and training use different OpenAI crawlers. OAI-SearchBot is the one that feeds ChatGPT search, and OpenAI’s crawler documentation is explicit: a site that opts out of OAI-SearchBot is not shown in ChatGPT search answers. It adds that robots.txt changes can take about 24 hours to take effect, and recommends allowing requests from its published IP ranges.⁵5 Source: OpenAI, “Overview of OpenAI crawlers”, accessed Oct 2026. You can block GPTBot (training) and still allow OAI-SearchBot; they are separate controls. Our AI search visibility guide has the full crawler map for OpenAI, Anthropic, Perplexity and Google.

Check the robots.txt file, then check the layer most teams forget: CDN and firewall bot rules. A bot-protection setting that challenges unknown crawlers can block OAI-SearchBot without anyone noticing.

Serve content without JavaScript

Vercel and MERJ analyzed AI crawler traffic across Vercel’s network and reported in December 2024 that “none of the major AI crawlers currently render JavaScript.” ChatGPT’s crawler fetched JavaScript files but did not execute them.⁶6 Source: Vercel, “The rise of the AI crawler”, accessed Oct 2026. Published December 17, 2024, with MERJ. A page that builds its pricing table or product copy in the browser looks empty to OpenAI. The fix is server-side rendering, static generation or a prerendering layer for bots.

The rest of the technical checklist

Check	What good looks like	Why it matters for ChatGPT
robots.txt	OAI-SearchBot allowed; GPTBot a separate decision	Blocking the search crawler removes you from answers5
CDN and firewall	OpenAI’s published IP ranges allowed	Bot rules can block the crawler silently5
Rendering	Key copy present in the initial HTML	OpenAI’s crawlers do not execute JavaScript6
Bing indexing	Sitemap submitted in Bing Webmaster Tools; key pages indexed	Bing is one of ChatGPT’s search providers2
Canonicals and noindex	Self-referencing canonicals; no stray noindex on pages you want cited	A page excluded from indexes cannot be retrieved
Hreflang	Matches canonicals on multi-language sites	Stops the wrong language version being read
Structured data	Organization, Product or SoftwareApplication, FAQPage where they fit	Helps parsing; OpenAI has not named it a ranking factor

5. Steps 4 to 6: build the pages ChatGPT reads

With a month of tracking data and a crawlable site, you can decide what to publish.

Step 4: reverse-engineer the winners for each prompt

For every tracked prompt, sort the cited URLs by type: vendor product pages, publisher roundups, review-site profiles, community threads, documentation. Patterns show up quickly. Trade-blog roundups may own one prompt, G2 and Capterra profiles the next, and a pair of Reddit threads a third.

That sort answers the most practical question in this work: does the prompt need a page on your site, a mention on someone else’s page, or both? When it does need a page, study the winners. Note their subheadings, the questions their FAQs answer, how they describe each option, and where the comparison table sits.

Step 5: publish the formats that get cited

For “best X” and “X alternatives” prompts, list-style comparison pages are often the most-cited format. Reach AI, an SEO and AEO agency, says its citation analysis for a fraud-prevention client found list-format pages took a third of all citations (33%) across the tracked prompts, ahead of every other format.⁷7 Source: Reach AI, “Fraudio case study” (usereach.ai/customer-results/fraudio), accessed Oct 2026. Figures reported by the agency. On Google, the pages that usually rank are still product, feature and location pages. Plan for two content tracks, one per channel.

Three rules keep this from backfiring:

• Be fair. A roundup where the author wins every row looks like advertising to readers and to models alike. Name where you fall short and which competitor suits a buyer better.
• Make each page different. Every variant needs its own arguments, competitors and facts. A copied page with a new title is the scaled, low-value content Google’s spam policies target.
• Publish in batches. Ship a handful of pages, watch the tracking data, then do more of what worked.

Step 6: answer the fan-out queries on the page

To be named inside the answer, the model needs a clear sentence it can lift for each query it ran. Write for that:

• Put the answer in the first sentence of each section, under a heading that states the question.
• State facts plainly: price, plan limits, integrations, who the product suits and who it does not.
• Use tables for comparisons, and add an FAQ built from real buyer questions.
• Date every page visibly and update numbers the week they change. Stale pricing is one of the errors assistants repeat most about brands, and it usually traces back to an old page.

Keep the company description consistent across every page: positioning, ideal customer profile (ICP) and the three or four attributes you want associated with the brand. When your own pages disagree with each other, the model has no reason to repeat any of them.

6. Steps 7 and 8: earn mentions where ChatGPT already looks

Most sources that decide your category’s answers are not on your domain. A competitor’s own roundup will never put you first, so independent pages carry much of the recommendation.

Step 7: get listed on the pages ChatGPT cites

Use the tracking data from step 2 to rank external domains by how often ChatGPT cites them for your prompts. Then work down the list: ask for inclusion in existing roundups, offer data or a product trial to the author, and fix your profiles on the review sites that appear. Choose targets by citation frequency, not domain fame. A mid-sized trade blog cited every week is worth more than a large publisher the model never reads for your prompts.

Sponsored listings on third-party roundups are a legitimate option and many publishers sell them. Their effect is indirect, through the page ChatGPT reads, so buy only on sites your tracking shows it citing, and only where the comparison is even-handed. Paying for placement inside ChatGPT’s answer is not on offer. OpenAI began testing ads in the US on February 9, 2026, for logged-in adults on the Free and Go plans, and has since expanded the pilot to markets including Canada, the UK, Brazil and Japan. It says ads are labeled as sponsored, shown separately, and “do not influence the answers ChatGPT gives you.”⁸8 Source: OpenAI, “Testing ads in ChatGPT”, accessed Oct 2026. Originally published February 9, 2026; updates through August 11, 2026 list the expansion markets.

Step 8: join the community threads that shape answers

Reddit and Quora threads appear in AI answers more often than most teams expect. Reach reports that for Carro, a dropship commerce network, Reddit was the most cited domain across the prompts it tracked, ahead of every brand.⁹9 Source: Reach AI, “Carro case study” (usereach.ai/customer-results/carro), accessed Oct 2026. Figures reported by the agency for March to May 2026.

Pull the threads that keep appearing in your tracking data and reply where you can add something concrete: a configuration tip, a real price, an honest comparison. Read each subreddit’s rules first and say that you work for the company. Promotional replies get deleted or buried, and a deleted reply teaches the model nothing. If a forum’s rules leave little room, put the detailed version on a help center or community space you run and let the thread link to it.

7. Step 9: report on visibility and revenue together

Report four numbers per prompt and per assistant: mention rate (share of runs that name you), citation rate (share that link to you), average position in the list, and sentiment. Then connect them to revenue.

• Referrals. Per OpenAI’s publisher FAQ, links clicked in ChatGPT carry utm_source=chatgpt.com, which makes them visible in Google Analytics and similar tools.¹⁰10 Source: OpenAI Help Center, “Publishers and Developers FAQ”, accessed Oct 2026 via search result text. Build an AI referral channel in GA4 and track key events such as booked demos.
• Self-reported attribution. Add a “where did you hear about us” field to demo forms and log the answers in the CRM.
• Branded and direct demand. Many people who meet a brand in an answer later type the URL or search the name. Watch branded search impressions in Search Console and direct traffic alongside AI referrals.

Clean click attribution is the floor of the effect, not the whole of it. Reach reports that at ColdIQ, a go-to-market agency, organic and AI search influenced $506,000 in contract value over three months, while direct-traffic users nearly doubled over the same period.¹¹11 Source: Reach AI, “ColdIQ case study” (usereach.ai/customer-results/coldiq), accessed Oct 2026. Figures reported by the agency; direct-traffic users rose 93% over the comparison window. Treat any single vendor case study with care; the pattern of AI exposure showing up as direct and branded demand is the part worth copying.

ChatGPT and Google compared

	Google	ChatGPT
The target	Keywords and positions	Prompts and the queries the model writes for them2
What the user sees	A ranked list of pages	A written answer naming a few brands, with citations
Crawler	Googlebot renders JavaScript	OAI-SearchBot does not run it6
Pages that usually win	Product, feature and location pages	Comparisons, roundups and alternatives pages for “best X” prompts
Off-site signals	Links	Mentions on cited pages, review profiles and community threads
Stability	Positions move gradually	Answers vary run to run; trends need weeks4
What to report	Rankings, clicks, conversions	Mention and citation rate, AI referrals, branded demand, pipeline

Strong Google positions help, since ChatGPT’s queries run against search indexes. They are not enough on their own. A site can be strong in Google and absent from ChatGPT’s shortlists when the cited roundups and threads never mention it.

8. How smaller brands win, and the mistakes that stop them

Brand size counts for less in ChatGPT than on Google, because the model cites whichever pages best answer the queries it writes. That gives challengers a real opening if they build the comparison pages incumbents have not.

Reach’s published case study on Fraudio, an Amsterdam fraud-prevention company of about 30 people competing with Feedzai, SEON and Sift, is a useful example of the sequence. According to Reach, Fraudio built list-format comparison pages first, translated the best performers, bought listings on outside publishers and added schema to product pages. Between February and October 2026 it moved from #124 to #12 among brands AI engines mentioned in its category and became the second most cited domain in fraud prevention.⁷ These are the agency’s own figures, but the order of work matches everything above: research, access, pages, then mentions.

Running it in-house or with a specialist

Nine steps is a lot of work for one marketer. Most teams end up with one of three setups:

12 Source: Ahrefs, “Pricing”, accessed Oct 2026. List price; Brand Radar is an add-on. 13 Source: Semrush Knowledge Base, “AI Visibility Toolkit”, accessed Oct 2026. List price. 14 Source: Reach AI, “Pricing” (usereach.ai/pricing), accessed Oct 2026. List prices, according to the company.

Setup	Who does what	List price as of October 2026	Fits
Fully in-house	Your team samples prompts by hand and runs steps 3 to 9	Staff time only	Small prompt sets; teams testing whether ChatGPT matters for them
Tracker plus in-house execution	An SEO-suite add-on tracks prompts; your team builds pages and earns mentions	Ahrefs Brand Radar AI from $199/mo;12 Semrush AI Visibility Toolkit $99/mo13	Teams with writers and an outreach owner already in place
Specialist SEO and AEO agency	The agency runs tracking, technical fixes, pages, placements and community work	Reach AI: $8,000/mo for 100 tracked prompts, $15,000/mo for 25014	Teams that want the full sequence run for them

The middle option is the most common and the easiest to stall: the tracker reports the gaps every week and nobody is assigned to close them. If you go that way, name an owner for steps 4 to 8 before you buy.

If you hire help, judge any agency on whether it can show you its prompt tracking, explain why a given page or placement was chosen, and report results in pipeline terms. Reach AI, whose Fraudio work is described above, is built around that sequence. Its retainers bundle daily prompt tracking with page production, link and mention placements, and community replies on Reddit, Quora and G2, so the person reading the data is also the one acting on it.¹⁴ For a team without writers or an outreach function, that is the main advantage over buying a tracker alone.

Mistakes to avoid

• Measuring the wrong questions. Branded and vague category prompts flatter the dashboard.
• Judging from one answer. The next run can name different brands.⁴
• A firewall rule that blocks OAI-SearchBot. It drops you out of ChatGPT search with no visible error.⁵
• Content that only renders in the browser. OpenAI’s crawlers see an empty page.⁶
• Publishing generic AI-written pages at volume. One page with specific facts beats twenty interchangeable ones.
• Roundups that read as ads for their author. Buyers and models discount them.
• Ignoring third-party pages and threads. Fixing only your own site produces citations without mentions.
• Paying for a tracker nobody acts on. Measurement without someone assigned to steps 4 to 8 is a cost, not a program.
• Reporting ChatGPT clicks alone. It misses the branded and direct demand the channel creates.

9. The verdict

Ranking in ChatGPT is less mysterious than the industry makes it sound. OpenAI documents the gates: allow OAI-SearchBot, be readable without JavaScript, be findable through its search partners. Past those gates, the model repeats what the pages it reads say. Your job is to make sure the pages that win its queries, on your site and off it, describe you accurately and favorably, and then to measure long enough to know it worked.

Frequently asked questions

How do I get my website to show up in ChatGPT?

Allow OAI-SearchBot in robots.txt and in your firewall, make sure key content is in the initial HTML, and get indexed in Bing. Next, build pages for the searches ChatGPT runs behind your buyers’ questions and win places on the outside pages it keeps citing. Give each prompt a month of tracking before deciding what worked.

How fast can a brand start appearing in ChatGPT?

Crawl fixes take effect within about a day of a robots.txt change, per OpenAI.⁵ Changes in which brands get recommended usually take months, because new pages need to be indexed and third-party mentions take time to land. Plan for leading signals within weeks and a judgment on pipeline after three to six months.

Does ranking on Google help me rank in ChatGPT?

It helps without settling the outcome. The queries ChatGPT writes go to search providers, so pages that are indexed and rank well get read more.² Brands that rank well on Google can still be missing from the roundups and threads that ChatGPT cites for buying prompts.

Can I pay to rank in ChatGPT?

No. OpenAI says ads in ChatGPT are labeled, shown separately from the answer and do not influence it.⁸ You can pay for sponsored listings on third-party sites that ChatGPT cites, which affects answers only through the pages it reads.

Does schema markup help?

It helps search systems parse page type, company details and FAQs, but OpenAI has not named it a ranking factor for ChatGPT search. Add Organization, Product or SoftwareApplication and FAQPage markup where they fit, after crawl access, rendering and the content itself are in order.

Is ChatGPT worth the effort if it sends little traffic?

Usually, if your buyers research before they buy. Much of its effect shows up as direct visits, branded searches and self-reported attribution rather than tagged referrals. Count mentions, citations and pipeline; treat tagged clicks as a lower bound.

Sources and notes

1. Source: TechCrunch, “ChatGPT reaches 900M weekly active users”, accessed Oct 2026. Published February 27, 2026, reporting OpenAI’s announcement. ↩

2. Source: OpenAI Help Center, “Searching the web with ChatGPT”, accessed Oct 2026. Covers automatic search, query rewriting, search providers and site eligibility. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

3. Source: Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview”, accessed Oct 2026. Covers Copilot and Bing AI answers, not ChatGPT. ↩

4. Source: SparkToro, “NEW Research: AIs are highly inconsistent when recommending brands or products”, accessed Oct 2026. Published January 28, 2026; 2,961 runs; not peer reviewed; co-author works at an AI tracking vendor. ↩ ↩² ↩³

5. Source: OpenAI, “Overview of OpenAI crawlers”, accessed Oct 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵

6. Source: Vercel, “The rise of the AI crawler”, accessed Oct 2026. Published December 17, 2024, with MERJ. ↩ ↩² ↩³ ↩⁴

7. Source: Reach AI, “Fraudio case study” (usereach.ai/customer-results/fraudio), accessed Oct 2026. Figures reported by the agency. ↩ ↩²

8. Source: OpenAI, “Testing ads in ChatGPT”, accessed Oct 2026. Originally published February 9, 2026; updates through August 11, 2026 list the expansion markets. ↩ ↩² ↩³

9. Source: Reach AI, “Carro case study” (usereach.ai/customer-results/carro), accessed Oct 2026. Figures reported by the agency for March to May 2026. ↩

10. Source: OpenAI Help Center, “Publishers and Developers FAQ”, accessed Oct 2026 via search result text. ↩

11. Source: Reach AI, “ColdIQ case study” (usereach.ai/customer-results/coldiq), accessed Oct 2026. Figures reported by the agency; direct-traffic users rose 93% over the comparison window. ↩

12. Source: Ahrefs, “Pricing”, accessed Oct 2026. List price; Brand Radar is an add-on. ↩

13. Source: Semrush Knowledge Base, “AI Visibility Toolkit”, accessed Oct 2026. List price. ↩

14. Source: Reach AI, “Pricing” (usereach.ai/pricing), accessed Oct 2026. List prices, according to the company. ↩ ↩²