ChatGPT recommends businesses in one of two ways. Without a web search, it answers from what the model learned in training. With a search, which it can run on its own when a question needs current or local facts, it rewrites the question into search queries, reads the results and names a few businesses.
To be in the search version: let OAI-SearchBot crawl your site, publish pages that answer the exact question with real facts, and get onto the lists, directories and review sites ChatGPT reads. OpenAI says "placement is not guaranteed", so measure progress on a fixed list of buyer questions.
How ChatGPT decides, with and without search
Without a search
- The answer comes from the model's training.
- GPTBot is the crawler that collects content that may be used for training.
- Facts can be out of date: old prices, a branch that closed, last year's offer.
- You cannot update this quickly. Businesses written about widely for years have an edge.
With a search
- ChatGPT "rewrites your query into one or more targeted queries" and sends them to search providers. The privacy policies it links for them are Microsoft's and Shopify's.
- It may add the user's general location from their IP address, and saved memories.
- After the first results, it "may send additional, more specific queries."
- It answers with citations. Pages that OAI-SearchBot can crawl are eligible.
OpenAI's own example shows how personal this gets. A user whose memory says they are vegan and live in San Francisco asks for nearby restaurants, and ChatGPT may search "good vegan restaurants San Francisco". In Dubai, the same "best physiotherapist near me" question can become a search for Dubai Marina from one phone and for Al Barsha from another. One screenshot from your own phone proves very little. We check in clean sessions and save every answer.
The three OpenAI crawlers and your robots.txt
robots.txt is a small text file at the root of your site that tells crawlers what they may read. OpenAI runs three agents, and they do different jobs:
| Crawler | What OpenAI says it does | Follows robots.txt? | If you block it |
|---|---|---|---|
| OAI-SearchBot | "used to surface websites in search results in ChatGPT's search features" | Yes | Your site "will not be shown in ChatGPT search answers" |
| GPTBot | "used to crawl content that may be used in training our generative AI foundation models" | Yes | Your content should not be used in training. It does not remove you from search |
| ChatGPT-User | Used "for certain user actions in ChatGPT and Custom GPTs" | "robots.txt rules may not apply" | Little effect, because a user asked for the page |
Most businesses should allow all three. If you want out of training but in search, block GPTBot and allow OAI-SearchBot. The two lines that matter most:
User-agent: OAI-SearchBotAllow: /
Two traps we check on every site:
- The firewall overrides the file. OpenAI says to confirm that your host or CDN "allows traffic from OpenAI's published searchbot IP addresses." Since 1 July 2025, Cloudflare has asked every new domain whether to allow AI crawlers. A robots.txt that says Allow means nothing if the firewall says no. Google makes the same point for its own AI features: crawling must be allowed "by any CDN or hosting infrastructure". Your server logs show whether OAI-SearchBot is actually getting through.
- Facts locked in images or PDFs. Put prices, hours, areas and services in plain text on the page. Google gives the same advice for its AI features ("making sure that important content is available in textual form"), and a price inside a picture gives any crawler nothing to quote.
Changes are not instant either. OpenAI says "it can take ~24 hours from a site's robots.txt update" for its systems to adjust.
Where ChatGPT's sources come from
For a "who should I use in Dubai" question, the pages we see ChatGPT cite in our checks fall into five groups:
| Source type | Examples | What you can do |
|---|---|---|
| Your own site | Service pages, price pages, location pages | Answer the exact question, with numbers |
| "Best of" and Top 10 lists | Blog posts that rank local businesses | Get onto honest lists that accept submissions. Never buy a spot |
| Directories and listings | Industry directories and local listings. For agencies, that means sites like Clutch, GoodFirms and Semrush Agencies | Complete profiles with the same facts as your site |
| Reviews | Review sites and platforms in your sector | A steady flow of recent reviews, and replies with facts |
| Forums, video and social posts | Reddit threads, YouTube videos, LinkedIn posts | Real participation by real people from your business |
Two more details matter for local businesses. OpenAI says ChatGPT may share a user's general location with "trusted third-party providers" for things like "nearby business listings", so your listings outside your own site feed local answers. And our team sees ChatGPT reach for Reddit, YouTube and LinkedIn mostly when the question is long and detailed. For comparison, Pew found Wikipedia, YouTube and Reddit were the three most common sources in Google's AI summaries.
Why "best X in Dubai" answers are built from Top 10 lists
Follow one question through the loop. A buyer asks ChatGPT for the best AI SEO agency in Dubai. ChatGPT turns that into a search, and the search returns lists. In a web search we ran on 21 October 2026 for "best AI SEO agency in Dubai", 7 of the first 10 results were "best" or "Top 10" lists. The three we opened each ranked their own publisher first or second. ChatGPT reads those lists and summarises them, so its answer repeats them, self-ranking included.
Check your own category the same way: search the question your buyers ask and count how many results are lists. Then three lessons follow:
- For "best" questions, the lists decide the answer, not your homepage. If you are on none of them, a great website will not save you.
- Specific questions are where you win. Real ChatGPT answers name The Setup Collective first for "Where can I get Christmas family photos taken in Dubai?" and Hype first for "Where can I buy good school backpacks for kids in Dubai?". "Cheapest UAE free zone" answers led with ANCFZ. None of these needed a Top 10 list.
- Shortcuts backfire. We don't write fake "independent" lists, buy placements or post negative content about rivals. Google's guide says "seeking inauthentic 'mentions'" is not "as helpful as it might seem", and a buyer who opens the source can see who wrote it.
What you can and can't control
You can control
- Whether crawlers can reach your site
- What your pages say, and how plainly
- Whether your facts match everywhere
- Your directory profiles and Google Business Profile
- Asking real customers for reviews, and replying
- Real Arabic pages
- Tagging your links so you can count leads
You can't control
- What the model learned in training
- How ChatGPT rewrites a question
- The buyer's location, memory or settings
- Which of several good pages it picks
- Answers changing from one run to the next
- Model updates
- Buying a place in the answer: OpenAI says ads "do not influence the answers"
Checklist: getting recommended by ChatGPT
- Allow OAI-SearchBot in robots.txt. Allow GPTBot too, unless you object to training.
- Check your firewall or CDN lets OpenAI's published IP ranges through, and confirm OAI-SearchBot visits in your server logs.
- Ask ChatGPT your top 10 buyer questions in a fresh chat, signed out (OpenAI says signed-out users can use web search too). Save every answer and the links it cites.
- For each question you lose, open the cited pages. Those are the sources you need to be on.
- Write one page for each question you lose, with price, timing, area, who it is for and who it is not for.
- Put those facts in plain text, not only in images or PDFs.
- Make your name, address, phone, hours and prices identical on your site, Google Business Profile, socials and directories.
- Create or complete your profiles on the directories in your industry.
- Ask to be included on honest "best of" lists that accept submissions.
- Ask recent happy customers for reviews, and reply to negative ones with facts.
- Publish Arabic pages for Arabic buyers, on their own URLs.
- Verify your site in Bing Webmaster Tools and submit your sitemap. OpenAI's search partners include Microsoft, and Bing's AI Performance report shows which of your pages Copilot cites and the "grounding queries" behind them.
- Tag your WhatsApp buttons and watch for
utm_source=chatgpt.comin your analytics. - Re-check the same questions every month, and change one thing at a time so you know what worked.
How to see ChatGPT traffic in GA4 or your own analytics
- Open the AI Assistant channel. In GA4, go to Reports, Acquisition, Traffic acquisition, and group by Session default channel group. Google sets the medium to "ai-assistant" when the referrer matches its list of AI assistants, which includes ChatGPT, Gemini, DeepSeek, Copilot and Grok.
- Also filter by source. On adjoltz.com, ChatGPT visitors arrive with
utm_source=chatgpt.combut no referrer, so they can miss the AI Assistant channel. Add a filter where Session source contains chatgpt.com. - Look at landing pages. The pages ChatGPT sends people to are the pages it cites. That list tells you what is working.
- Count actions, not visits. Mark WhatsApp taps, calls and form sends as key events, then break them down by source.
- Know the October 2026 gap. From 9 October 2026, many ChatGPT links arrived with no tag and no referrer, so visits showed as direct. Adobe's Cedric Huesler reported the tag back on 15 October (Search Engine Roundtable). If your ChatGPT numbers fell that week while direct traffic rose, that was the cause.
- Ask every lead how they found you. It catches what no tag can: a copied phone number, a screenshot forwarded by a friend.
Our own numbers, for scale: 4 ChatGPT visits to adjoltz.com on 20 and 21 October 2026. One landed on an Arabic page and one of the four tapped WhatsApp.
ChatGPT vs Claude: what our team has noticed
This section is our team's observation from testing, not a published fact. Models change often, so treat it as a snapshot from October 2026.
In our tests, ChatGPT goes along with the framing of a question faster. Ask "Is X the best clinic in JVC?" and it tends to lean towards yes and build the case. Claude tends to weigh more outside sources before it agrees. OpenAI has corrected this kind of behaviour before: in April 2025 it rolled back a GPT-4o update that was "overly flattering or agreeable".
Some differences are documented. Anthropic's subprocessor list names Brave Search for Claude's web search, so Claude searches a different index from ChatGPT's partners, and a page can be easy to find in one and hard in the other. Claude has its own crawlers: ClaudeBot for training, Claude-User for pages a user asks it to open, and Claude-SearchBot for search. Anthropic says blocking Claude-SearchBot "may reduce your site's visibility" in its search results.
What we do with this: we never judge visibility from a leading question. We use neutral wording ("Who are good options for X in Y?"), and we check the follow-up questions as well as the first answer. Scale matters too. In September 2026, ChatGPT carried 80.99% of AI chatbot referral traffic in the UAE and Claude 1.23% (Statcounter), which is why our plans track ChatGPT and Gemini (Google AI).