How to Optimize for ChatGPT Search: Access, Tracking and Content

To appear in ChatGPT search, let OpenAI's search crawler, OAI-SearchBot, reach your pages, then publish clear, specific, well-sourced answers to the questions your buyers ask. OpenAI documents the access rules. It does not publish a ranking formula, so the rest is good editing, honest evidence and patient testing.

ChatGPT is where many people now ask a question in full sentences and expect a direct answer with links. If your business is the answer, you want to be named and linked. This guide covers only what is specific to ChatGPT search: the crawlers OpenAI documents, how to check access, how to track visits, how to test your visibility and what to publish. For the platform-agnostic process, read how to get cited by AI search engines. For the concepts, see our GEO guide.

What ChatGPT search is, in one paragraph

ChatGPT search is the part of ChatGPT that looks up current information on the web and writes an answer that links to the pages it used. It differs from a classic search engine in one important way: it returns a written answer first and a handful of links second, so being cited matters more than being ranked tenth. OpenAI says any public website can appear in ChatGPT search, and that to be discovered and cited you must not block its search crawler.

The OpenAI bots, and what each one does

OpenAI publishes a crawler overview that separates its bots by job. Treat them separately, because you can allow some and block others. The table summarises the current documentation. Check the source before you change anything, as the list and the user-agent versions can change.

User agent	Purpose as OpenAI describes it	Should you allow it?
OAI-SearchBot	Surfaces websites in ChatGPT's search features. Sites that opt out will not be shown in ChatGPT search answers, though they can still appear as navigational links	Yes, if you want to appear in ChatGPT search
GPTBot	Crawls content that may be used to train OpenAI's generative AI foundation models	Your policy decision. It is independent of search visibility
ChatGPT-User	Visits a page when a user asks ChatGPT or a custom GPT something. Not used to crawl automatically, and robots.txt rules may not apply because a user initiated it	Usually leave unblocked at the firewall
OAI-AdsBot	Validates the safety of pages submitted as ads on ChatGPT, and does not train models	Allow it if you plan to advertise there

The key line in OpenAI's documentation is that each robots.txt setting is independent. A site can allow OAI-SearchBot to appear in results and disallow GPTBot so its content is not used for training. OpenAI also says that for search results it can take about 24 hours after a robots.txt update for its systems to adjust. Source: OpenAI's overview of OpenAI crawlers and its Publishers and Developers FAQ.

Check access in four places

Most "we are invisible in ChatGPT" problems are access problems. Check these four layers in order.

1. robots.txt

Open yoursite.com/robots.txt and look for rules that apply to OAI-SearchBot or to all user agents. A blanket Disallow: / under User-agent: * blocks it. A copied template from a security plugin can do this without you noticing. A simple policy that allows search and blocks training looks like this. Adjust it to your own decision.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

2. Your firewall or CDN

Bot protection in a content delivery network or web application firewall can block a crawler even when robots.txt allows it. OpenAI publishes the IP ranges for its bots as JSON files, and recommends allowing OAI-SearchBot from them. Ask your host or developer to check the firewall logs for blocked requests from the OAI-SearchBot user agent. If you find them, add an allow rule that combines the user agent with the published IP ranges, so that an impostor using the name alone is not let through.

3. Page-level controls

Check that important pages are not set to noindex and do not hide their content behind a login, a cookie wall that blocks the text or a script that only loads the content after interaction. A crawler needs to see the words in the response.

4. Server logs

The proof is in your logs. Search for the OAI-SearchBot user agent and confirm it receives 200 responses on your key pages. If the bot never appears at all, either it has not found your pages yet or something upstream blocks it before the request reaches your logs.

Track ChatGPT visits in your analytics

OpenAI says its referral links automatically include the parameter utm_source=chatgpt.com, which lets you track visits from ChatGPT search in tools such as Google Analytics. In practice that means you can build a simple view without any special tool.

1. In Google Analytics 4, open the traffic acquisition report and set the primary dimension to session source.
2. Search for chatgpt.com. Save the view or create a segment for it.
3. Add the landing page as a secondary dimension to see which pages ChatGPT sends people to.
4. Add your conversion events to see whether those visitors enquire or buy.

Expect small numbers at first and treat them as a trend. Also remember that some people see your name in an answer and visit later by typing your address or searching for your brand, which will show up as direct or organic traffic. A rise in branded search after you start appearing in answers is a useful secondary signal.

How to test your visibility in ChatGPT

There is no ChatGPT equivalent of Search Console, so testing is manual. A careful routine gives useful signals despite that.

1. Write your question list. Use 20 to 40 questions a real buyer would ask, in full sentences, covering early, middle and late stages. Include "best [service] in [city]", "how much does [service] cost" and comparisons.
2. Use search deliberately. Make sure web search is being used for the answer, and keep the same method every month.
3. Use a fresh conversation each time. Earlier messages and saved memory can change answers.
4. Record the results. For each question note whether you are named, whether you are linked, which sources are cited, and what the answer says about you.
5. Repeat the same question a couple of times. Answers vary. Look at patterns across questions, not one result.
6. Keep a log of the date and the wording. Without it you cannot see change.

Be careful with conclusions. A single answer that does not mention you does not mean you are invisible, and a single mention does not mean you have won. You are looking for a direction over months.

What to publish so ChatGPT has something good to cite

OpenAI does not publish ranking factors for search, so what follows is practitioner judgement based on how answers are assembled from web pages, with Google's published guidance as a useful cross-check. Google's guide for its own AI features stresses unique, non-commodity content that adds a point of view or first-hand experience. The same instinct serves any engine that has to choose between near-identical pages.

Answer the question a buyer actually asks

Take your top ten sales questions and give each one its own page or clearly headed section. Put a direct answer first, with the conditions that change it.

Publish real prices or ranges, and what moves them

Cost is one of the most common questions people ask an assistant, and most sites dodge it. A page that gives an honest range, the factors that move it and a date for when it was last reviewed is useful to readers and easy for a system to quote.

Write the comparison pages your buyers need

People ask assistants to compare options. A fair "X vs Y" page with a table, clear criteria and a plain recommendation for each kind of buyer is a strong format. Be honest about where a competitor is better. Pages that read as pure sales copy are less likely to be trusted by readers and models alike.

Add first-hand evidence

Show what you have actually done: process photos, anonymised before-and-after numbers with your method stated, the mistakes you have seen. Do not invent case studies. Name clients only with their agreement.

Keep facts current and dated

If a price, rule or feature changes, update the page and show the real update date. Stale pages are a risk where answers depend on current information.

Make the page easy to read as text

Use real headings, short paragraphs, real HTML tables and lists, and descriptive link text. Keep the important text in the page itself, not inside images.

Brand and reputation signals

A model that must recommend a business tends to rely on what the wider web says about it. Reviews on trusted platforms, accurate listings, mentions in credible publications and consistent business details help it describe you correctly. This is reasonable inference, not a published OpenAI rule. It also comes with a caution. Google's guide says seeking inauthentic mentions is not as helpful as it might seem, and the principle carries over. Earn mentions by being genuinely useful and by asking real customers for honest reviews. Never buy reviews or mass-produce low-quality placements.

What not to spend time on

• Special files. Neither the OpenAI crawler documentation nor Google's guide treats an llms.txt file as a requirement. Google says its Search ignores them. Publishing one is optional and should not be sold as a fix.
• Tricks aimed at one model version. They stop working when the model changes.
• Prompt injection or hidden text. Hiding instructions in your pages for AI is deceptive, can break search policies and risks being ignored or penalised.
• Dozens of near-duplicate pages targeting every phrasing of a question. They dilute your site and, for Google, can breach scaled content rules.
• Guarantees. Nobody can guarantee a citation.

ChatGPT search vs Google: what differs for you

	Google AI features	ChatGPT search
Crawler to allow	Googlebot	OAI-SearchBot
Your own control panel	Search Console, including a Generative AI performance report	None. Use analytics and logs
Referral tracking	Impressions in the Generative AI report, clicks pooled in overall performance data	utm_source=chatgpt.com on referral links, per OpenAI
Opt out of AI features	Search Console setting for generative AI features	Disallow OAI-SearchBot. You can still appear as a navigational link
Official optimisation guide	Yes, from Google Search Central	Crawler documentation and a publishers FAQ. No ranking guide

Reading your server logs for OAI-SearchBot

Logs are the only direct evidence that the crawler reaches you. Ask your developer or host for access to the raw access logs, then look for three things.

1. Presence. Search the logs for the string OAI-SearchBot. If you find entries, the bot is reaching your server. If you find none over several weeks, either it has not found your pages or something upstream is blocking it.
2. Response codes. Check the status code on each request. 200 is healthy. 403 usually means a firewall rule, 429 means rate limiting, 404 means the page is missing and 5xx points to server errors.
3. Coverage. List which URLs it requested. If it fetches the home page but none of your service pages, check your internal linking and sitemap.

On a typical Linux server a simple search is grep "OAI-SearchBot" access.log | awk '{print $9, $7}' | sort | uniq -c | sort -rn | head -50, which lists the most requested paths with their status codes. Field positions differ by log format, so ask your developer to adjust it. Because user agent strings can be copied by anyone, confirm important findings against the IP ranges OpenAI publishes before drawing conclusions.

A prompt panel template you can copy

Use a simple spreadsheet with one row per question per month. These columns are enough.

Column	Example entry
Month	October 2026
Funnel stage	Late: cost question
Question (exact wording)	How much does a website for a small accountancy firm cost in the UK?
Run number	1 of 2
Brand named?	No
Brand linked?	No
Domains cited	Three agency sites and one directory
Answer summary	Gave a range and listed three cost drivers
Gap spotted	Our pricing page has no range and no date
Action	Add dated range and drivers to pricing page

The "gap spotted" and "action" columns turn the panel from a report card into a to-do list. Without them, you collect data and change nothing.

Example questions by funnel stage

For a service business, a balanced list of 30 questions might include these kinds. The wording is invented for illustration.

Stage	Example question	Page that should own it
Early	What is the difference between a website redesign and a rebuild?	Explainer
Early	Do small businesses need an SEO audit?	Explainer
Middle	How do I choose a web design agency?	Buying guide
Middle	Shopify or WooCommerce for a small shop?	Comparison
Late	How much does a 10-page business website cost?	Pricing page
Late	How long does a website project take?	Process page
Brand	What does [your company] do and where is it based?	About page
Brand	Is [your company] any good? Reviews	Reviews and case studies
Local	Best web design agency in Manchester for a small business	Location service page

Three common situations and what to do

Situation 1: you never appear and the bot never shows in your logs

This points to access. Check robots.txt, then the firewall and CDN bot settings, then whether your pages are discoverable through internal links and a sitemap. Fix the block, wait at least a few days after OpenAI's stated 24-hour window and look at the logs again.

Situation 2: the bot visits, but competitors are cited and you are not

Access is fine, so the question is content. Open the cited pages and compare. Does each answer the question more directly than yours? Do they give a price, a date or a named source that you lack? Rewrite your page to match the usefulness, then add something original that they do not have.

Situation 3: you are mentioned, but the details are wrong

The system is drawing from outdated or conflicting sources. Correct your own pages first, then update directories and profiles that carry the old information. Publish a clear, current page for the fact in question and re-test after a month.

Who should own this in your business

Assign three roles even if one person fills all of them. A technical owner checks robots.txt, the firewall and logs. A content owner maintains the question list and the pages. A reporting owner runs the monthly panel and shares what changed. Put a recurring monthly reminder in a shared calendar, because this work fails most often by being forgotten, not by being hard.

A 30-day plan

Week	Action	Done when
1	Check robots.txt, firewall and logs for OAI-SearchBot. Decide your GPTBot policy	Logs show OAI-SearchBot receiving 200s on key pages
2	Create the question list, run the first test in fresh conversations and set up the chatgpt.com view in GA4	A baseline sheet and a saved analytics view
3	Rewrite or create the three pages that match your most valuable questions, with ranges, comparisons and evidence	Three pages published with real dates and sources
4	Correct inaccurate third-party listings, ask real customers for reviews, re-run the test	A before-and-after comparison and a plan for the next month

Deciding your training policy

Because search and training are separate settings, you have four possible combinations. There is no single right answer, so choose with your business model in mind.

Choice	Effect as OpenAI describes it	Suits
Allow OAI-SearchBot, allow GPTBot	You can appear in ChatGPT search, and content may be used for model training	Businesses focused on reach, with content they are happy to share widely
Allow OAI-SearchBot, disallow GPTBot	You can appear in search, and signal that content should not be used for training	Most businesses that want visibility but control over reuse
Disallow OAI-SearchBot, allow GPTBot	Not shown in ChatGPT search answers, though content may be used for training	Rarely sensible
Disallow both	Not shown in search answers, and a signal against training	Publishers with paid content or strict licensing needs

Review the choice once a year. Policies, products and your own strategy all change. Also record who made the decision and why, so a later website update does not undo it by accident.

Special situations

You sell online

Accurate product data matters wherever you are listed. Keep titles, prices, availability and delivery information correct on your own pages, and in any feeds you submit to platforms. Write size, fit and compatibility answers where shoppers can find them.

You are a local business

Keep your business profile and listings consistent in name, address, phone, services and hours. Publish pages for the areas you serve with real, local detail. Answer pricing and timing questions in plain terms.

You publish information for a living

Decide your policy on training and search deliberately, as the two are separate switches. Think about what a visitor gets from you that an answer cannot supply, such as tools, data, community or depth, and build for that.

You advertise

OpenAI documents a separate bot, OAI-AdsBot, that checks the landing pages of ads submitted on ChatGPT. If you plan to advertise there, make sure it is not blocked. We discuss the wider shift in ads inside AI answers.

Common mistakes

• Blocking OAI-SearchBot through a blanket robots.txt rule or a firewall setting, then wondering why you never appear.
• Confusing GPTBot with the search crawler, and blocking the wrong one.
• Judging success from a single answer.
• Publishing thin pages with generic advice that any site could write.
• Ignoring analytics, so visits from ChatGPT go unseen.
• Leaving inaccurate facts about your business on third-party sites.
• Paying for a package that promises guaranteed placement.

Where we can help

We are a digital marketing agency in Manchester, UK and Mumbai, India. Our ChatGPT SEO services start with the access audit and a baseline test, then move to content and reputation work. You can also contact us and we will tell you plainly what we find.