How to get recommended by ChatGPT: a practical plan

ChatGPT recommends brands in two ways. It can answer from what its model learned during training, or it can search the web and build an answer from the pages it finds. OpenAI says ChatGPT may search automatically when a question would benefit from current information, and that when it uses search partners it usually rewrites the question into one or more targeted queries.^[2] For a prompt like "best payroll software for a 20-person company in Canada", that means ChatGPT is effectively running several web searches and summarizing what comes back.

Which pages make the cut is not published in detail. OpenAI's publisher FAQ says ChatGPT ranks search results using multiple factors intended to surface relevant, reliable information, and that placement is not guaranteed.^[3] Treat anyone who claims to know the exact formula with caution.

Three OpenAI user agents matter to site owners. Each is documented separately, and each does a different job.^[1]

OpenAI also says robots.txt changes take about 24 hours to take effect, and it publishes IP address lists for each crawler so hosts and firewalls can allow them.^[1] You can test your own setup in a few seconds with the AI crawler checker.

A brand gets mentioned when the pages ChatGPT reads describe it clearly and in the right context. Your own site is one input. Independent sources that compare, review or list businesses like yours are usually just as important, because an unbranded question ("which CRM suits a small agency?") is answered from pages that are about the category, not about you.

• Crawlability. If OAI-SearchBot is blocked by robots.txt, a CDN rule or a firewall, your pages cannot be cited. OpenAI asks site owners to allow its published search crawler IP addresses at the host or CDN level too.^[3]
• Third-party coverage. Comparison articles, "best of" lists, industry directories, trade press and forum threads are the pages that answer category questions. If you are absent from them, you are often absent from the answer.
• Reviews. For shopping questions, OpenAI says ChatGPT considers things like price and reviews, uses structured product data from first-party and third-party providers, and may summarize reviews from public websites.^[4]
• Entity clarity. Use one consistent name, say plainly what you do, who it is for and where you operate, and keep those facts identical across your site, profiles and listings. Ambiguity makes it harder for any system to connect a mention to you.
• Answerable pages. Pricing, specifications, service areas, integrations and honest comparisons give a summarizing model concrete facts to quote. Vague marketing copy gives it nothing to work with.

Research supports the idea that content changes can move AI visibility, within limits. In a benchmark study, Aggarwal and colleagues reported that targeted content changes raised visibility in generative engine answers by up to 40%, with results varying a lot by subject area.^[7] That is a lab result on a research engine, not a promise about ChatGPT, but it is a reason to keep testing rather than guessing.

The plan below runs from the quickest technical checks to the slower work of earning coverage. Most businesses can finish steps 1 to 4 in a week. Steps 5 to 8 are ongoing.

1. Write down 20 to 40 real questions your buyers would ask without naming you, in each language and country you sell in. A dental clinic in Melbourne, a B2B software company in Berlin and an online furniture store in Mexico City will each have very different lists. Add a few branded questions too, but keep them separate.
2. Check crawler access. Confirm robots.txt allows OAI-SearchBot, and that your CDN, bot-protection settings and firewall do the same. Cloudflare, for example, announced in July 2025 that it was changing its default to block AI crawlers, so sites behind it should review their settings.^[8]
3. Fix your entity basics. Make the homepage and About page say exactly what you are, who you serve and where. Align your name, category and location across directories, review sites and social profiles.
4. Publish the pages people ask about. Pricing (or how pricing works), service areas, use cases, integrations, comparisons with named alternatives, and FAQs written in the words buyers use. Keep each page specific and current.
5. Find the pages ChatGPT already cites for your questions. Run your list, note the third-party domains that keep appearing, and treat them as your outreach map.
6. Earn a place on those pages. Ask for inclusion in relevant comparisons, get listed in the directories your industry uses, pitch data or expertise to trade publications, and answer questions in the communities where buyers ask them. Disclose any relationship.
7. Encourage genuine reviews on the platforms that matter in your market and category. Never buy or fabricate them.
8. Re-measure monthly with the same question set, and change one thing at a time where you can, so you learn what moved the result.

Build pages that answer a buying question with facts, and earn mentions on pages that compare options. The table pairs common question types with the page that tends to answer them.

Segment matters. A SaaS company usually wins through comparisons and integrations, an online store through product data and reviews, and a local business through directories and its map listing. The SaaS and e-commerce pages go deeper on each.

Shortcuts mostly waste time. A few of them can also damage trust in your brand.

• Paying for answers. There is no paid placement inside ChatGPT's responses, according to OpenAI.^[5]
• Fake reviews and planted mentions. Beyond the ethical problem, they tend to get removed and can do lasting damage to trust in your brand.
• Blocking `GPTBot` and assuming you are out of ChatGPT. Training and search are controlled by different crawlers.^[1] Decide each one on purpose.
• Treating llms.txt as a ranking lever. It is optional hygiene. See the llms.txt guide for what it does and does not do.
• Judging from one screenshot. The same prompt can produce different brands minutes apart, so a single check is not evidence of progress or decline.

Measure the share of answers that mention you, across a fixed set of unbranded questions, asked many times. ChatGPT's answers vary run to run, and OpenAI's model release notes show the default model changing during 2026, including a May update that changed when ChatGPT decides to search the web.^[6] A single result cannot separate your progress from that background movement.

• Use several paraphrases of each question, not one exact wording.
• Keep the country and language fixed per measurement, and label them.
• Record the number of answers behind every percentage, plus a range.
• Report branded questions separately. ChatGPT naming you when asked about you by name is expected, not a win.
• Track referral traffic too. OpenAI says ChatGPT adds utm_source=chatgpt.com to referral links, so it shows up in analytics tools.^[3]

Glossy Data samples ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity and Claude weekly across paraphrased questions in 49 countries, and every figure carries its sample size and a 90% range. The methodology explains why, and why AI answers change walks through the statistics. For the ChatGPT-specific details, see ChatGPT tracking.

Technical fixes can show up within days. OpenAI says robots.txt changes take about a day to apply.^[1] New pages need to be discovered and then picked for relevant searches, which can take weeks. Third-party coverage is the slowest part: earning a place on comparison lists, collecting reviews and building a press record usually takes months.