How to get cited by ChatGPT (and recommended in its answers)

ChatGPT names a business in one of two ways. Either the model already knows you from what it learned in training, or it searches the web while it writes the answer and cites the pages it finds. Each route needs different work from you.

In the free version, memory does most of the work. Ahrefs found that only 14.4% of 66.6 million free ChatGPT answers from August and September 2026 ran a web search at all, which means that for most questions the model answers from what it already knows about you, or does not know.

Buying questions behave differently. When we asked ChatGPT about home appliances in Egyptian Arabic in October 2026, 58 of 60 answers cited at least one live source.

When we asked for local businesses, with questions like “what’s the best lab in Nasr City?”, 287 of the 358 businesses it picked were backed by map data.

This guide walks through both routes, the seven steps that feed them, the shortcuts to avoid and how to check whether any of it worked. You will also see what AI answers were worth in leads to one of our clients in Saudi Arabia. Every figure carries its source and date.

The seven steps at a glance

Here is the whole method in one table. Steps 1 to 4 apply to every business, step 5 matters most if you serve a local area, and steps 6 and 7 decide whether the work keeps paying off and whether you can prove it moved anything. Each step is explained below with the evidence behind it.

How does ChatGPT decide which businesses to name?

It draws on two sources. The first is what the model learned in training, which OpenAI’s GPTBot crawler helps collect; the second is live search, where ChatGPT reads pages fetched by OAI-SearchBot and cites them with links beside the answer. One answer can use both. Being strong on one route does not put you on the other.

We read the Ahrefs study through a write-up by Relevant Audience, not the original report. Answers that did search ran about 1.6 sub-queries each.

The study covers one model snapshot and US-only citation data, and it notes that the paid and reasoning models search more than the free one.

You can see the memory route in our Saudi health panel. On 10 September 2026, one Arabic answer named four hospitals in Jeddah but cited only two; the other two had nothing beside them to click, because they came from what the model already knew.

What do the two routes mean for an Arab business?

If you run a local business, the search route matters most, and it runs through your map listing. If you sell a product brand, memory carries more weight, and memory follows how much people search for you and write about you. Our three studies, run in Egypt and Saudi Arabia in September and October 2026, point the same way.

• Local services. We asked ChatGPT 100 questions in Egyptian Arabic on 13 October 2026. Map data was the strongest evidence behind 80.2% of its 358 recommendations, and the business’s own website was the evidence for 12.6%. The details are in our study of local recommendations in Egypt.
• Product brands. Across 25 appliance brands and 60 questions in Egyptian Arabic, asked on 13 October 2026, Egyptian search demand and ChatGPT recommendations correlated at 0.91 by rank. The details are in our study of appliance brands in Egypt.
• Question language. When we asked the same 60 questions in English on 14 October, ChatGPT gave made-in-Egypt brands 24.7% of recommendation slots, against 33.0% in the Arabic repeat.
• Health questions. We asked ChatGPT 29 Saudi health questions in both languages. For Arabic informational questions, 97.5% of its 40 Arabic citations went to institutions, including CDC and NHS pages that exist only in English. On four “best doctor or hospital” questions it cited no institution at all. The details are in our study of Arabic and English citations.

Before you spend money on either route, work out which one your buyers’ questions take.

Do AI answers actually bring in customers?

They can. Delta Medical Labs in Saudi Arabia, a VOCTOS client, recorded 728 form and booking leads from SEO and AI search in the 30 days to 6 November 2026. Of those, 313, or 43%, came from AI assistants. Delta stopped Google Ads in autumn 2026 and now relies on organic search and AI.

Read that number with its limits. Delta’s tracking groups referrals from AI assistants together, and chatgpt.com sometimes shows up as a separate source, so we added it in. The 313 therefore covers AI assistants as a group, not ChatGPT alone.

Calls and WhatsApp chats are not in either number. It is also one client over one month, and we are the agency doing the work, so treat it as one case, not a benchmark.

The practical lesson is about the visit as much as the mention. Someone who arrives from an AI answer has already read a recommendation, so make the next step easy to find, whether that is a booking form, a phone number or a WhatsApp button.

You can read about the SEO and GEO work behind these numbers in the Delta Medical Labs case study.

Does ChatGPT search in English when you ask in Arabic?

Nobody has published that measurement for Arabic, and that includes us. Peec AI found that for non-English prompts in its data, ChatGPT added at least one English search in nearly 78% of cases, and 43% of all its sub-queries ran in English. Arabic is not among the languages Peec names, so for Arabic the question is still open.

Peec’s analysis, published in February 2026, covered more than 10 million prompts and counted only users whose country matched the prompt language. Turkish switched to English most often, at 94%, and Spanish least, at 66%. Peec sells tracking software, so read it as a vendor study.

Our own studies recorded the sources ChatGPT cited, not the searches it ran, so they cannot settle this.

A University of Toronto preprint by Chen and colleagues points the other way. It found GPT “completely swaps its domain ecosystem to favor local-language sources” when the prompt language changes.

That study did not test Arabic either. Under both readings the advice is the same: keep one set of facts about your business in Arabic and in English, so a search finds the same story in whichever language it runs.

Step 1: Is OAI-SearchBot allowed to crawl your site?

This is the only requirement OpenAI documents. OpenAI’s crawler overview says sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers, though can still appear as navigational links.” GPTBot, the training crawler, is a separate setting, so you can block training and still show up in search.

OpenAI’s publishers FAQ puts it plainly: “make sure you aren’t blocking OAI-SearchBot.” A minimal robots.txt looks like this:

User-agent: OAI-SearchBot
Allow: /

# Only if you want to opt out of model training:
User-agent: GPTBot
Disallow: /

• Allow OpenAI’s published IP ranges in your firewall and CDN as well. OpenAI recommends this on the same page, because a bot rule at that level blocks the crawler even when robots.txt lets it in.
• Use noindex for pages you want kept out of search. The crawler has to be allowed in to read the tag.
• Give a robots.txt change about 24 hours to show up in search.
• Ignore ChatGPT-User here. OpenAI says it “is not used to determine whether content may appear in Search.”

Evidence strength: strong. This comes from OpenAI’s own pages, read on 27 October 2026. But crawler access is a gate, and it will not move you up on its own.

In our appliance study, none of the robots.txt files we read blocked OpenAI, including those of brands ChatGPT never named.

Neither OpenAI page asks for an llms.txt file. Our guide to robots.txt for AI crawlers covers the crawler settings in detail. If someone is trying to sell you an llms.txt file, read our guide to what an llms.txt file does.

Step 2: Write your business facts in plain text, in both languages

When ChatGPT searches, it checks brands’ own sites for facts. Write your legal name, category, services, branches, price ranges and customer service number as plain text on your own pages, in Arabic and English, and keep them identical everywhere you appear. Comparison answers lean on other people’s sites, while factual answers lean on yours.

In the Ahrefs data, 6.8% of 15 million sub-queries used the site: operator. Of the site: searches aimed at about 160 popular domains, 75.9% targeted brand sites. For factual questions, 63.4% of citations went to official sites, against 46.4% for comparisons.

Evidence strength: moderate. We read it second-hand, and it covers only US prompts.

Our appliance data shows what happens when you leave the facts to others. In 3 of 10 answers about repairs and spare parts, ChatGPT cited third-party sites with a brand’s name in their domain, and one of them advertised a different Toshiba number from the official 19111 hotline.

If you do not publish your own Arabic service page, someone else’s page answers for you.

Put the answer near the top of each page, in plain sentences that still make sense when quoted on their own. A price or a hotline buried in the sixth paragraph helps nobody.

Then keep the same facts on your Google Business Profile, your LinkedIn page and any Wikidata entry you have, so every source tells the same story.

Structured data helps machines match the facts, as our guide to brand entities and knowledge panels explains, but treat it as housekeeping. Product markup did not separate recommended brands from missing ones in our appliance study.

Step 3: Get other sites to mention you, in Arabic and English

Mentions on other sites are the strongest correlate of AI visibility in the large studies we cite here, and they feed both routes. A shop page, a news story or a client’s announcement that names you is read by search today and can enter training later. Arabic coverage takes its own effort, separate from English.

• Ahrefs, 75,000 brands: web mentions correlated with visibility at 0.664 and backlinks at 0.218. This was measured in Google AI Overviews, not ChatGPT, and Ahrefs says correlation is not causation.
• Seer Interactive, about 10,000 GPT-4o answers: page-one Google rankings correlated with brand mentions at about 0.65 and backlinks at about 0.10.
• Chen and colleagues: earned media made up 93.5% and 95.1% of ChatGPT’s sources in two experiments. The paper is not yet peer reviewed.
• Muck Rack, May 2026, more than 25 million links: earned media was 84% of citations, journalism 27% and paid content 0.3%. We read it through a Muck Rack press release, and Muck Rack sells PR software.

Egypt shows the same pattern. In our appliance study, shop and dealer sites were cited in 31 of 60 answers, led by B.TECH and Jumia at 12 each, and brand sites in 27.

Only 20.2% of 208 recommendations came with the brand’s own site cited. The shops did the talking.

Evidence strength: moderate. The direction is consistent, and every result is a correlation.

In practice, start with the shops that sell you and the press your buyers read.

In Egypt that could be Daily News Egypt, Al-Mal or Amwal Al Ghad, in Saudi Arabia Argaam, and across the region Wamda or Magnitt. Our guide to digital PR for GEO covers how to pitch them, and our guide to Arabic GEO covers writing for Arabic answers.

Step 4: Keep reviews and directory listings current

ChatGPT writes review counts beside the businesses it recommends, and in our Egypt data a thin review count was the warning it attached most often. Directory and booking listings get cited too, especially for doctors and clinics. Keep the profiles that already show up for your category complete, consistent and honestly reviewed.

In our local study, 233 of 358 recommendations had a rating written into the sentence, and of the 22 picks that came with a caution, 7 were about having few reviews. A low count did not rule anyone out, though. 42 single-branch businesses were recommended with fewer than 50 reviews, one of them with only two.

ChatGPT cited Egyptian directories such as Egyfinder and Dalil 140, and not one Facebook page in 200 answers.

In Saudi health answers, directories and booking platforms took 10.6% of ChatGPT’s Arabic citations, against 3.3% of its English ones. So if you run a clinic in Riyadh or Jeddah, your booking platform profile is more likely to be cited in an Arabic answer than in an English one.

For agencies and B2B firms the evidence is weaker. Hall found Clutch behind 84.5% of review-platform citations for digital services in ChatGPT, while DerivateX found review sites at 1% across 40 software categories. Both are vendor studies of mostly US prompts.

Want to see which of these you are missing? Show me what ChatGPT says about my business, and we will run your buyers’ questions in Arabic and English.

Step 5: For a local business, is your map listing doing the selling?

In Egypt, your map listing is what ChatGPT shows. Across 100 local questions on 13 October 2026, 289 of 358 recommended businesses appeared as place cards with a name, category, star rating and Directions button, and for 287 of them map data was the strongest evidence behind the pick. OpenAI has not said where that card data comes from. Keep every map listing you control accurate.

1. The listing name. ChatGPT used it as the heading, branch included, for example “4level1, Serag Mall branch”.
2. The listing category. The card prints it under the name, so it is the first thing a user reads about you.
3. Rating and review count. ChatGPT quoted both inside the recommendation.
4. The website button. Only 131 of the 289 cards carried a Website button.

If you run a chain, you need one accurate listing per branch, because ChatGPT names the branch. A pharmacy with branches in Nasr City and Maadi needs both listed properly, and so does a clinic group spread across Riyadh.

Real estate developers are the exception: only 10 of their 34 picks rested on map data, and 14 came with property portals and buyer guides instead.

Evidence strength: moderate. This is our own study, in one country, logged out, over two days.

Step 6: Publish original data only you can supply

Original data gives ChatGPT a source it cannot find anywhere else, and it gives journalists a reason to name you. The Ahrefs tracking suggests ChatGPT moved toward primary sources during 2026. In our Saudi panel, the two languages agreed on a source mainly when one page owned the answer for everybody.

Between March and September 2026, Reddit’s share of US ChatGPT answers fell from 19.1% to 2.3%, and PubMed rose from 64th to 4th place among cited domains. Muck Rack found that 56% of ChatGPT’s journalism citations were published within the past year.

Evidence strength: moderate. Both figures reached us second-hand, and there is no causal test behind either.

You do not need a research team for this. A dated price index for your category, your average delivery times in Cairo or Riyadh, or the questions your customers ask most will do, as long as you add a short note on how you collected it.

A translated copy of a foreign guide does the opposite, because it gives ChatGPT no reason to pick you over the original. We publish each of our studies with every raw answer, so anyone can recount them.

Step 7: How do you measure whether ChatGPT recommends you?

Use a fixed set of buyer questions, ask them several times in Arabic and in English, and count how often you are named. One screenshot proves very little. When we asked the same 100 Egyptian questions a day apart, ChatGPT recommended only 58.7% of the same businesses again, and it kept the same first pick in 52 answers.

• Write the questions the way your buyers type them: Egyptian or Gulf colloquial, formal Arabic and English. A shopper in Cairo starts with “eih ahsan” and one in Riyadh with “wesh afdal”, though both are asking which is best. In our appliance study, formal Arabic moved the lists about as much as asking twice, and English moved them further.
• Ask each question more than once, logged out, in a new chat, and note the date. Report the share of answers that name you, not a rank.
• Track referrals from chatgpt.com. OpenAI says ChatGPT adds utm_source=chatgpt.com to referral links, but many links lost the tag between about 8 and 16 October 2026, so mark that week in GA4, as we explain in our write-up of the October tagging gap.
• Watch branded search as well. People who hear your name from ChatGPT may go and search for it, so your branded impressions in Google Search Console are worth tracking next to AI referrals.
• Ask every new lead how they found you. Deals closed on WhatsApp or by phone never reach your analytics, which is exactly the gap in the Delta numbers above.

Our guide to measuring AI visibility sets out the full method. The three studies above publish their questions and every answer, so you can copy the design for your own category.

What should you never do to get into ChatGPT?

Three shortcuts carry real risk: ranking yourself first in your own “best of” list, buying or gating reviews, and hiding instructions for AI in your pages. The evidence against each beats the evidence for any of them, and two can cost you listings and trust you already have. A fourth, paying for a guaranteed spot, mostly wastes money.

Self-ranking “best of” lists

In June 2026, Lily Ray checked about 100 US software queries. The brand that ranked itself first in its own list was left out of the Google AI Overview recommendation 69% of the time, as Search Engine Land reported.

By September, AI Overviews cited 38% fewer of these lists and left their owner out 83% of the time, according to PPC Land. Evidence strength: moderate. This is Google, not ChatGPT.

Fake, paid or gated reviews

Google’s Business Profile policy on fake engagement lets Google remove reviews, block new ones, unpublish existing ones and show a public notice on your profile.

In our Egypt data, ChatGPT quoted the review count beside its picks, so a restricted profile damages the exact line your buyers read. Evidence strength: strong. This is Google’s own policy.

Hidden prompt text

In December 2024, The Guardian found that hidden text made ChatGPT search give positive verdicts even where the visible reviews were negative, as Search Engine Journal reported. It worked in a test. That does not make it safe.

Hidden text has long broken Google’s spam policies, and Google confirmed five spam updates between January 2025 and October 2026, all listed in our tracker of Google algorithm updates. If a journalist finds hidden instructions on your site, the damage to your name is obvious.

Buying links for “AI authority” earns little too: backlinks correlated at 0.218 for Ahrefs and about 0.10 for Seer.

Paying for a guaranteed spot

Nobody can promise you a place in ChatGPT’s answers. The only requirement OpenAI documents is crawler access, and as Step 7 shows, the same question can return a different list the next day.

If an agency offers that guarantee, put our 25 questions to ask an SEO or GEO agency to it first. Other popular claims that do not hold up are collected in what ChatGPT optimization guides get wrong.

My read: the gap is in memory, not in the crawler

After three studies this autumn, my view is that crawl settings rarely keep Arab businesses out of ChatGPT. The bigger gap is that too few independent pages, especially in Arabic, describe them at all. Fix step 1 in an afternoon, then put most of your effort into steps 3, 4 and 5.

The appliance data is what convinced me. The five brands ChatGPT never named in 120 Egyptian Arabic answers were among the six least searched in Egypt, and their robots.txt files did not explain it.

Haier is the case I keep coming back to. It has 9,880 monthly searches for its appliances in Egypt and an English-only Egyptian site, and it got 3 recommendations against 7 for Kiriazi, which has similar demand.

I cannot prove the site is the reason. I can say demand alone did not carry it.

Questions business owners ask us about ChatGPT

These are eight worries business owners raise when they first check what ChatGPT says about them, written the way owners put them. Each answer points back to the evidence above. Where we have no data we say so, because a confident guess about ChatGPT is easy to sell and hard to check.

We rank on Google’s first page and ChatGPT still recommends a competitor.

Ranking helps, but it does not decide.

Seer found a correlation of about 0.65 between page-one rankings and ChatGPT mentions, which leaves a lot unexplained, and only 14.4% of free answers in the Ahrefs sample searched at all. Compare your competitor’s reviews, listings and mentions with yours first.

Did blocking GPTBot remove us from ChatGPT?

Not from search. Training is a separate switch.

OpenAI treats the two crawlers independently, and only OAI-SearchBot decides whether your pages appear in ChatGPT search answers. Blocking GPTBot tells OpenAI not to train on your content. If you want to stay in search, keep OAI-SearchBot allowed.

We are not sure whether to write the page in Arabic or English.

Write both, with the same facts.

In our appliance study, English questions gave Egyptian brands 24.7% of recommendation slots against 33.0% in Arabic. Each language rewards different sources, so a single version leaves half your buyers to someone else.

Someone told us we need an llms.txt file to appear in ChatGPT.

Not according to OpenAI.

Neither its crawler overview nor its publishers FAQ asks sites for an llms.txt file. The documented requirement is OAI-SearchBot access. The file does no harm, and our guide to llms.txt files explains what it does, but do not pay for it as a ChatGPT fix.

How long until ChatGPT starts recommending us?

For search, OpenAI says robots.txt changes take about 24 hours, and new pages still have to be found. For memory, change waits for new models, and OpenAI publishes no schedule.

We have no before-and-after data that supports a timeline, and anyone who quotes you one is guessing.

Is traffic from ChatGPT worth the effort, or is it a vanity number?

For one of our clients it has been worth it.

Delta Medical Labs in Saudi Arabia, a VOCTOS client, recorded 728 form and booking leads from SEO and AI search in the 30 days to 6 November 2026, and 313 of them, or 43%, came from AI assistants.

That figure groups all AI assistants, not ChatGPT alone, and leaves out calls and WhatsApp chats. One client is not a benchmark, so measure your own.

Our competitor shows up in ChatGPT ads and we do not.

Ads sit outside the recommendation list.

In our Egypt local study, 35 of 100 ChatGPT answers carried a sponsored unit labelled Ad beside the organic picks, and no advertiser was also named as a recommendation in the same answer. In that data, paying bought the ad slot, not a place on the list.

ChatGPT recommended us yesterday and not today.

That is normal. Each answer is one draw.

In our repeat run, 58.7% of businesses came back the next day, and chains came back more often than single branches, 68.0% against 55.2%. Judge where you stand by the share of many answers that name you, over weeks, not by one chat.

The better question is which route you are missing

“How do I get into ChatGPT?” is the wrong first question. Ask which of the two routes your buyers’ questions take, and which one you are missing from. A dental clinic in Jeddah and an appliance brand in Cairo need different first steps, even though both want the same sentence with their name in it.

If you serve a local area and your buyers ask “best X in Y”, start with steps 5 and 4, then confirm step 1.

If you sell a product, a professional service or B2B, start with steps 3 and 6, and finish steps 1 and 2 in week one.

Either way, run step 7 before you begin, or you will not know what changed.

This guide is part of our AI search series. The other engines each have their own guide:

• Getting cited by Perplexity
• Getting cited by Gemini
• Getting cited by Copilot
• Getting cited by Grok and DeepSeek

If you would rather have us do this for you, our ChatGPT optimization service page explains how we work and what we measure. To start from your own data instead, run my buyers’ questions through ChatGPT.