# Why Does ChatGPT Recommend My Competitors and Not Me?

> ChatGPT recommends your competitors if it can't find, read or trust your company. Here is how to see which one it is, and the first fix to make.

By Michael Zacs · How AI Picks · Published Nov 16, 2026 · Last updated Nov 16, 2026
Source: https://exibran.com/blog/why-chatgpt-recommends-my-competitors

## Quick answer

ChatGPT recommends your competitors when it can find, read and trust more about them than about you: a clear description, mentions on other sites, and pages that answer the buyer's exact question. Ask your five buyer questions five times each on three AI apps, count who is mentioned and who is recommended, then fix the biggest gap first.

## Key takeaways

- It is rarely a product problem. One B2B software company was mentioned in 0 of 96 AI answers, yet AI described it well when asked by name.
- One answer proves nothing. On average, only 30% of brands stayed visible in back-to-back AI responses, so count across repeated runs.
- Letting the crawler in is not enough. A software company that allows crawlers and has an llms.txt file still scored 0 of 96.
- Nobody can honestly promise a placement. What you can get is a counted baseline and a clear first fix.

On 7 and 8 Nov 2026, I asked AI apps the questions one B2B software company's buyers ask. That is 96 answers. The company was mentioned in none of them.

When I typed its brand name, AI described it well. It just never came up on its own. I counted every answer, because a guess does not tell you what to fix.

If ChatGPT recommends your competitors and not you, you are not imagining it. It is also not a verdict on your product. It usually means the AI could not find, read or trust enough about you. All of that can be fixed.

People call this work AI visibility, generative engine optimization (GEO), answer engine optimization (AEO) or AI search optimization. I use plain words here.

**In this guide, you'll understand:**

- Why ChatGPT picks your competitors, in the reasons AI itself gave
- How to count who it recommends, in one sitting, with a free sheet
- What to fix first, with copy-and-paste text and a 7-day plan
- What other founders tried, and what I am doing at Exibran

> **Short on Time?**
>
> Check your robots.txt first (see "Can ChatGPT Reach and Read My Website?"), then follow the 7-day plan. Everything else in this guide supports those two things.

*Image: ChatGPT Recommends a Brand*

*ChatGPT Recommends a Brand*

## Why Does ChatGPT Recommend My Competitors Instead of Me?

ChatGPT recommends your competitors when it finds clearer, more consistent, and more widely confirmed information about them than about you.

I call this **FRT** - Found, Read, and Trust.

1. Found - Does your site clearly answer your customers' questions? Is your product being discussed across different sources?
1. Read - Can ChatGPT easily access your site? Is anything blocking your site?
1. Trust - Do you truly deliver on what your product promises? Do people say good things about your product? Are your product details consistent everywhere?

*Image: Three things ChatGPT needs before it recommends a company: find it, read it and trust it*

*Your rival gave ChatGPT all three. You are missing at least one.*

I asked 11 AI answers this exact question (ChatGPT, Perplexity, Google AI Mode, Google AI Overviews, and Gemini). Three reasons came up in every one:

| Reason AI Gave | Answers (of 11) | What to Check on Your Side |
| --- | --- | --- |
| Clear, consistent description of what you do | 11 | Open your homepage and pricing page. Does the first screen say what the product does, for which industry and company size, and what it costs? |
| Online presence and mentions | 11 | Search "best [your category] tools" on Google and open the first 10 results. How many of those pages mention you? |
| Third-party validation (reviews, roundups, press) | 11 | Are you on G2, Capterra and the "best of" articles that those 10 results show? |
| Technical access for AI crawlers | 9 | Does yourdomain.com/robots.txt block OAI-SearchBot? |
| Strong traditional SEO | 2 (Gemini only) | Do you rank on Google's first page for "best [your category] for [your buyer]"? |
| Price in plain text | 1 (Gemini only) | Does your pricing page print a price, even "from $X a month"? |

These are what AI says about itself, so treat them as leads to check, not as a formula. OpenAI publishes no ranking formula.

A good Google rank does not settle it either. One study of 278 buyer prompts found that [58% of Google's top-10 pages were also cited by ChatGPT](https://www.position.digital/blog/chatgpt-ranking-factors/). That leaves 42% that were not.

**Do this now:** write down the five questions your buyers ask before they pick a provider, in their words. You will use them in a minute.

### What the AI Said Instead

In that 96-answer check, 41 answers recommended a brand and 55 recommended no one. Dedicated comparison pages were advised in 42 answers. Not one answer mentioned the company.

That list is the to-do list: say on your homepage who the product is for, print the price on your pricing page, and put named customer results on a public case study page.

*Image: Dot grid of 96 AI answers: 55 recommended no one, 41 recommended a brand, 0 mentioned the company*

*96 answers, 0 mentions of the company. Most answers recommended no one.*

## Why Does ChatGPT Know My Brand but Not Recommend It?

ChatGPT can know your name and still leave you out, because knowing your brand name and recommending it are different outcomes.

I score every AI answer into one of four, and the fix is different for each.

| Outcome | What the Answer Does | First Thing to Try |
| --- | --- | --- |
| Not mentioned | Your brand is absent and your site is not used | Check robots.txt for OAI-SearchBot, then get listed on the pages the answers cite, such as G2, Capterra and roundup articles |
| Cited but not mentioned | Your page is a source, your name is not in the answer | Put your company name and one plain description in the cited page's first sentence |
| Mentioned but not recommended | You are in the list, rivals get the pick | Read the reason the answer gives for the pick (price, company size, an integration), then state that same thing on your homepage and pricing page |
| Recommended | The answer points the buyer to you | Keep it: re-run the same questions monthly |

*Image: Four steps from not mentioned to cited, mentioned and recommended*

*Each step needs a different fix.*

> If your site clearly says what you solve and who it’s for, and ChatGPT still doesn’t recommend you, look at the sites it cites. Get listed on them, by outreach or by joining the conversation there. ChatGPT wants several sources to agree on the same thing about your brand before it recommends you.

### What Happens When the Name Is in the Question

In my 96-answer check of a B2B software company, the company was mentioned in 0 of 96 answers to eight buyer questions.

Two extra questions had its name in them, and all 24 answers to those mentioned it. When someone already knew the name, AI described the product well. Buyers who had not heard of it never got that far.

On one question, that company's own site was cited in 3 of 12 answers and never mentioned in any of them. That is the "cited but not mentioned" outcome. It is cheap to test: open each cited page and check whether its first sentence names the company.

A January 2026 preprint found the same gap. [Amit Prakash Sharma tested 112 startups](https://arxiv.org/abs/2601.00912): ChatGPT recognised 99.4% of them when the query had the product's name, and surfaced 3.3% in open category questions.

It also found that products scoring high on a GEO checklist were no more likely to show up. So do not start with a checklist. Start with the count below.

**Do this now:** on the same app, ask "What is [your company]?" and then "Best [your category] for [your buyer]".

If the first describes you well and the second leaves you out, you are in the "not mentioned" row.

Typing your own brand name into ChatGPT hides the problem, because the name pulls you into the answer.

## How Do I Check Which Competitors ChatGPT Recommends?

There are two ways you can check which of your competitors ChatGPT recommends:

### 1. Check it Manually For Free

First, open an incognito/private browser window or log out of your personal account so past chat history doesn’t skew the results.

Ask your five buyer questions five times each on ChatGPT, Gemini and Perplexity, log every answer, then count. That gives 75 answers. Exibran's free check does the same thing for you, and the steps below are the same ones.

#### Step 1: Write Five Questions the Way Your Buyers Ask Them

Use the shapes real buyers type, not your product name. My 405-answer test of one software category used these:

| Question Shape | Example |
| --- | --- |
| Best in a category for a type of company | "Best AI visibility tracking tools for B2B SaaS companies" |
| Rival versus rival | "[Tool A] vs [Tool B]: which is better for a B2B SaaS team?" |
| Alternatives to a rival | "Cheaper alternatives to [Tool A] for AI visibility monitoring" |
| Price | "How much does an AI visibility tracking tool cost?" |
| Doubt | "Are AI visibility trackers accurate, or is the data noise?" |

Leave out any question with your own brand name in it. I count those separately.

### Step 2: Ask Each Question Five Times on Three AI Apps

Ask ChatGPT, Gemini and Perplexity each question five times, in a new chat every time. Five questions, five runs and three apps give 75 answers.

Running 75 answers through a tool took me about 25 minutes. By hand it takes longer. If you are short on time, start with one app: that is 25 answers, and it still shows you the pattern.

### Step 3: Log Every Answer in One Sheet

Copy this header row into a spreadsheet and fill one row per answer:

| Q# | App | Run | Brands Mentioned, in Order | Brand Recommended | You Mentioned? (Y/N) | Pages Cited |
| --- | --- | --- | --- | --- | --- | --- |

### Step 4: Open the Pages the Answers Cited

In ChatGPT, open the Sources panel under the answer. Links in ChatGPT answers usually end in `utm_source=chatgpt.com`, which marks them as coming from ChatGPT.

For each cited page, write down who is listed on it and whether you are. In my 96-answer check, one marketing-software site was cited in 24 of 96 answers (25%), and the company I checked was not listed on it. That became the first item in that report's action plan.

![ChatGPT citations](../../assets/images/posts/why-chatgpt-recommends-my-competitors/chatgpt-citations.png)

### Step 5: Count the Result

Count four things: how often you were mentioned, how often you were recommended, how often your site was cited, and which rival was recommended most. Keep them separate. A rise in one does not mean a rise in the others.

## Why Do ChatGPT's Answers Change Every Time I Ask?

Each ChatGPT answer is built fresh, so one answer tells you almost nothing and a count across repeated runs tells you a lot. AirOps found that on average [only 30% of brands stayed visible in back-to-back responses](https://www.airops.com/report/how-citations-mentions-impact-visibility). That is why I report a range, not a single number.

### How to Read a Count of Zero

A zero is still a result, and it comes with a range. Two of my checks counted zero mentions. The range for the true rate narrows as answers increase:

| Check | Answers | Mentioned | Range for the True Rate |
| --- | --- | --- | --- |
| 8 questions, 4 apps, 3 runs each (7 to 8 Nov 2026) | 96 | 0 | 0 to 3.8% |
| 27 questions, 5 apps, 3 runs each (3 Nov 2026) | 405 | 0 | 0 to 0.9% |

*Image: Chart showing a zero in 96 answers allows a true rate up to 3.8 percent, and a zero in 405 answers allows up to 0.9 percent*

*The more answers you count, the smaller the range a zero leaves open.*

So a zero out of 96 means the real rate is probably under about 4%, not exactly zero. The winners have a range too.

In the 405-answer test, the top three tools were each picked in about 10% of answers, and 212 of the 405 answers recommended no one at all. It is a few names sharing a small slice, not a leaderboard.

### Why Other AI Apps Answer Differently

ChatGPT, Gemini, Perplexity and Google's AI Overviews are separate systems, and each company describes how its own works.

A result on one app does not carry over to another. For every answer, add columns to your sheet for the app, the date, your country and whether web search was on. Count ChatGPT, Gemini and Perplexity on their own, and do not blend them into one score.

### What Not to Do

Do not report one screenshot to your team as proof of anything. If you have time for one thing, run the same five questions five times and report the range.

Write the date beside every count, because answers also shift when the model behind them changes.

## Can ChatGPT Reach and Read My Website?

ChatGPT can only use your pages in search answers if its search crawler, OAI-SearchBot, is allowed in.

[OpenAI's crawler documentation](https://developers.openai.com/api/docs/bots) says a site that opts out will not appear in ChatGPT search answers, though it can still show up as a link when someone asks for the site by name.

### How to Check Your Site's robots.txt

1. Open yourdomain.com/robots.txt in a browser.
1. Look for a line that says `User-agent: OAI-SearchBot` followed by `Disallow: /`. That blocks ChatGPT's search crawler.
1. Look for `User-agent: *` followed by `Disallow: /`. That blocks every crawler that has no rule of its own.
1. If neither is there, you are not blocking it through robots.txt.

OpenAI also says OAI-SearchBot and GPTBot are set independently, so you can allow search and still opt out of training. Changes can take about 24 hours to be picked up.

I ran this check on a GEO agency's site on 16 Nov 2026. OAI-SearchBot has its own rule with `Allow: /`, so ChatGPT's search crawler can read the site.

![Allowing ChatGPT to access your site](../../assets/images/posts/why-chatgpt-recommends-my-competitors/allowing-chatgpt-to-access-your-site.png)

### Check Your Firewall and CDN Too

A robots.txt that welcomes the crawler does nothing if a firewall or CDN setting blocks it first.

If you use Cloudflare, open [Security Settings](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots), scroll down to **Bot traffic**, select **AI Crawler and Agent access policies,** and find the AI bot controls.

![security settings on Cloudflare to ensure you don't block ChatGPT from your site](../../assets/images/posts/why-chatgpt-recommends-my-competitors/security-settings-on-cloudflare.png)

ChatGPT's search crawler counts as a search bot, so make sure the Search setting is on Allow and **Save**. You can still block AI training separately.

![Allow ChaGPT to crawl your site on Cloudflare](../../assets/images/posts/why-chatgpt-recommends-my-competitors/allow-chagpt-to-crawl-your-site-on-cloudflare.png)

### Can ChatGPT Read Text That Loads With JavaScript?

Probably not, so assume it cannot. A Vercel and MERJ study of crawler traffic from December 2024 found that [OpenAI's search and user crawlers did not render JavaScript](https://examples.vercel.com/blog/the-rise-of-the-ai-crawler). The data is old, so treat it as a reason to test, not as a rule.

Test your own page in two steps:

1. Open your page, right-click and choose View Page Source (not Inspect), then search for your product description.
1. If the text is there, a crawler that does not run scripts can read it. If you see only an empty shell and script tags, ask your developer to serve the key text in the HTML.

Prefer not to check by hand? The free [GEO Checker for SaaS](/tools/geo-checker) reads your home page and key pages the way AI crawlers do, without running JavaScript, and tells you which ones have almost no text in the HTML.

### Three More Technical Checks

OpenAI says your host or CDN must also allow its search crawler's published IP addresses, not only robots.txt.

Ask your developer to confirm it. Then check two ordinary things: type site:yourdomain.com/pricing into Google to see that your key pages are indexed, and search the page source for rel="canonical" to see that each page's canonical tag points to itself. These are standard search checks.

I have no ChatGPT test showing they change recommendations, so treat them as a clean-up, not a fix.

### Check That the Crawler Actually Visited

Your server keeps a diary of every visitor, called a log. Ask your developer, or look in your host's panel ([Cloudflare](https://developers.cloudflare.com/ai-crawl-control/features/analyze-ai-traffic/), [Vercel](https://vercel.com/docs/ai-gateway/observability-and-spend/logs), AWS or your hosting dashboard), to search the last 30 days of logs for the name OAI-SearchBot.

A result code of 200 means the crawler read the page. A 403 means something blocked it, usually a firewall. Nothing at all means it has not visited yet. Peec AI's server-log guide says the healthiest pattern is to see all three OpenAI bots: GPTBot, OAI-SearchBot and ChatGPT-User.

### What About Bing?

Some AI answers I collected told me to fix Bing first. That advice looks out of date. Peec AI's September 2026 study found ChatGPT now runs its own search index, with Google and several other providers feeding in, and Bing showing up mainly in Deep Research, the slower mode where ChatGPT works on a topic for several minutes.

That is a researcher's reading of ChatGPT's network traffic, not an OpenAI statement.

So the first job is still letting OAI-SearchBot in. Bing Webmaster Tools is free and reports citations in Microsoft's own AI features, and I found no Microsoft statement that it covers ChatGPT.

### Why This Is the Start and Not the Fix

The software company that scored 0 of 96 and 0 of 405 already allows crawlers in, and it points to an llms.txt file. Access was not its problem. If your robots.txt is clean, the next job is the fixes below.

## What Should I Fix First to Get ChatGPT to Recommend My Business?

Fix the things that cost nothing and unblock the rest first: let the crawler in, say one plain thing about your product, then find the pages ChatGPT already trusts and get onto them honestly.

| # | Fix | Done When |
| --- | --- | --- |
| 1 | Let ChatGPT's search crawler in | robots.txt and firewall do not block OAI-SearchBot |
| 2 | Write one plain description: what it does, for which industry and company size, and the price | The same sentence, name and price appear on your homepage, pricing page and every profile |
| 3 | Open every page the answers cited and list who is on it | You have a list of cited pages and who each one recommends |
| 4 | Get listed honestly on those pages | You are on G2, Capterra and at least two of the roundup articles from step 3 |
| 5 | Build pages for the exact questions: alternatives, comparisons, reviews, price | One page per question buyers asked, with the answer in its first paragraph |
| 6 | Put proof on your own pages: a case studies page with customer results and the customer's name, an integrations page, and a security page | A buyer can find a result with the customer's name on it without asking sales |
| 7 | Re-run the same questions and count | Same questions, same apps, same runs, new date |

### Write One Plain Description (Copy and Paste)

Fill this in once, then paste it on your homepage, pricing page and About page, and on your LinkedIn company page, Crunchbase profile, G2 and Capterra listings, Product Hunt page and X bio: "[Company] is [what it is] for [industry] companies with [size] staff. It helps them [main job]. Pricing starts at [price] a [month or year]. It is not for [who it does not fit]."

Keep the order and the wording identical everywhere. In my 96-answer check, AI could not state one company's price, and two sources gave it as $19 and $76 a month. One repeated sentence removes that gap.

If your pricing page says only "Contact Sales", a buyer who asks for the "best tool under $100 a month" cannot be matched to you, while a rival that prints "from $49 a month" can. One AI answer in my research made this point. I have not tested it, so treat it as a reason to print at least a starting price.

Use one company name too. In that check, ChatGPT mixed the company up with a different company that has a similar name, and recommended the wrong one. Add a plain line to your About page saying what you are not, for example "[Company] is not [similar name]."

### Name the Exact Category You Belong To

A company that calls itself "an AI business" gives a buyer's question nothing to match.

One that says "claims automation software for insurers" does. Several AI answers in my research said vague slogans make it hard to match a company to what a buyer asked.

Do this now: type your category into Google, read the "People also ask" box and the titles of the first 10 results, and pick the phrase that repeats most. Then use it in your homepage headline, page title, About page and every profile.

### Let the Crawler In (Copy and Paste)

If your robots.txt blocks ChatGPT's search crawler, replace the block with these two lines. Leave any GPTBot rule as it is, because blocking training is a separate setting.

```
User-agent: OAI-SearchBot
Allow: /
```

### Build the Pages Buyers Search For

One founder who got users from ChatGPT [built four kinds of pages](https://www.indiehackers.com/post/i-got-my-saas-recommended-1-by-chatgpt-here-s-exactly-what-i-did-XvbabP7l8707Rxbz2UUc): "best of" lists, reviews of a competitor, "A vs B" comparisons and "alternatives to A" pages. He made five to ten of each and improved on the pages ChatGPT already cited. Start with the rival your buyers type most often, and build one alternatives page in this order:

*Image: Five ring gauges showing what 9 cited pages share, such as text in the HTML and a questions section in 9 of 9*

*Most cited pages answered buyer questions and were updated in the last year.*

1. In the first paragraph, say who the page is for, what you cost, and one case where the rival is the better pick.
1. Add a table: you against the rival on price, company size, setup time and who should choose which.
1. Add one customer result with the customer's name on it, or a clear placeholder until you have permission.
1. Add 3 to 5 questions a buyer would ask, in their words, each answered in the first sentence.
1. Link it from your pricing page and your homepage.
1. Show a date, and update the page for real. 7 of the 9 pages ChatGPT's answers cited were updated within the last 12 months.

Then write one page on how to choose and one on what it costs. Say where you are not the right pick. One AI answer I read called a vendor's own comparison "vendor positioning, not independent proof", and honesty is what separates yours from that.

### Get Onto a Page ChatGPT Already Cites (Email Template)

From your sheet, take the cited pages that list your rivals and not you. Send the editor this, with no payment offered:

> Subject: Missing from your [list name]?
> 
> Hi [name], I read "[page title]" while comparing [category] tools. It lists [A], [B] and [C]. We make [Company]: [your one plain description]. Our pricing is here [link], and here is a result from [customer, with permission] [link]. If we fit your criteria, would you add us? If not, what would you need to see first?

Send five a week and log each reply in your sheet. The only account with numbers that I found is one founder who emailed 25 to 30 cited pages and got 7 link swaps, so treat this as a test and count what happens.

### Keep Your Reviews Recent

Two AI answers in my research listed stale or thin reviews as a reason a rival wins, and one cited page, Parvaly, lists a weak or stale review profile as its third cause.

I found no independent test that isolates review recency. The closest measured result comes from Trustpilot, which sells review software: in a May 2026 study of more than 800,000 AI answers, brands with an active review profile and regular replies were cited in 75.3% of answers, and the highest rate went to brands with more than 80 reviews.

Do this now: ask three customers who signed up this quarter to review you on G2 and Capterra, reply to every review, and check your Product Hunt page for old, unanswered comments.

In my 450 recorded answers, reddit.com and YouTube were cited more than G2 and Capterra, so do this alongside Reddit replies and a video, not instead of them.

### Publish Your Own Numbers

Other sites quote a company that has counts nobody else has, and several AI answers in my research said so.

A GEO agency's 405-answer test is why this guide can show counts instead of opinions.

Do this now: run your five buyer questions five times each, publish the result with the date as a blog post on your own site (for example "We asked ChatGPT 25 buyer questions: who it recommended"), even if the count is small and the result is zero.

### Your 7-Day Plan

*Image: Timeline of seven daily jobs from checking robots.txt to setting the re-check date*

*One small job a day. Day 7 sets your re-check date.*

| Day | Do This | Done When |
| --- | --- | --- |
| 1 | Read your robots.txt and add the two lines above if ChatGPT's search crawler is blocked | OAI-SearchBot is not blocked |
| 2 | Run the View Page Source test on your homepage and pricing page | Your product description is visible in the source |
| 3 | Write your one plain description and paste it on both pages | The same sentence and price appear on both |
| 4 | Ask your five buyer questions five times on three apps | Your sheet has 75 rows |
| 5 | Open every cited page and list who is on it and whether you are | You have a list of cited pages |
| 6 | Build your first alternatives page | It is live and linked from your pricing page |
| 7 | Send your first five emails and put the re-check date in your calendar | Re-check date set, such as 30 days out |

## What Have Other Founders Tried, and What Has a GEO Agency Seen?

Other founders have already tried most of the fixes above. Each row says what they did, what they reported and who measured it. Use the rows to pick what to try first. Every one is the founder's or a vendor's own test, so none of them is a promise.

| What Was Tried | Who Measured It | Result Reported |
| --- | --- | --- |
| Emailed 25 to 30 pages ChatGPT already cited, offered link swaps (7 agreed, 3 more paid $200 in total), then built list, review, comparison and alternatives pages | [Tom Han, a founder, on Indie Hackers](https://www.indiehackers.com/post/i-got-my-saas-recommended-1-by-chatgpt-here-s-exactly-what-i-did-XvbabP7l8707Rxbz2UUc), Nov 2026 | Not mentioned by ChatGPT in July 2026, then about 50 users a month from ChatGPT by November. Self-reported |
| 13 blog posts, one Reddit thread that got 5.3K views, and 6 directory listings | [A founder on Indie Hackers](https://www.indiehackers.com/post/i-asked-chatgpt-to-recommend-my-product-it-had-no-idea-it-existed-f30e86f4f5) | His product's AI visibility score went from 12 to 32 out of 100 in 11 days, on a score from a tool he built |
| Integration guides that mention his product (Zapier, n8n, Supabase), plus a community template | [A solo SaaS founder, on his own blog](https://rentierdigital.xyz/blog/ai-chatbot-saas-recommendations-seo) | ChatGPT mentioned his product in 0 of 15 prompts, then 4 of 15 two months later. Self-reported |
| Rebuilt a category page, then added community-sourced FAQs | [Product Hunt, in its own case study](https://producthunt.com/p/producthunt/case-study-how-product-hunt-can-improve-ai-visibility-in-2026) | ChatGPT citation rate of about 0.4% rose 10 times. They say they cannot tell how much was caused by the change |

I could not find an independently audited before-and-after for a B2B software company, so where that matters in this guide, I say so.

### What I Have Seen in My Checks

I ran two checks on one B2B software company in November 2026. Both counted zero mentions, and both use public buyer questions.

| Check | Date | Answers | Mentioned | Recommended | Recommended No One |
| --- | --- | --- | --- | --- | --- |
| A B2B software company | 7 to 8 Nov 2026 | 96 | 0 | 0 | 55 (41 of 96 recommended a brand) |
| The same company, wider test | 3 Nov 2026 | 405 | 0 | 0 | 212 |

In the 405-answer test, the most recommended rivals were each picked in about 10% of answers, and the company's own site was cited in none of them. The full method is in my [AI Pick Rate guide](/blog/ai-pick-rate).

### What I Am Doing About It at Exibran

Exibran is new, so I am doing this work in public. Here is the plan, and I will report what happens:

1. **High-intent buyer pages.** Comparison, alternatives and "how to choose" pages that answer the questions buyers ask when they are ready to pick.
1. **Showing up where AI reads.** Helpful replies and posts on Reddit, X, LinkedIn and Quora.
1. **Writing for other sites.** I am preparing outreach to write for sites AI already cites, so the pages it reads carry my name.
1. **One re-check.** After a month of consistent content, I run the same 75 answers again on the same three apps and publish the result here, whatever it is.

## Do llms.txt, Schema, Reviews, or Paid Mentions Work?

My research found no measured evidence that llms.txt, schema, review counts or paid mentions alone make ChatGPT recommend a company.

Only one fix is documented by OpenAI itself: letting its search crawler in. Each source below covers only the engine it names.

| Claim | What Can Responsibly Be Said | Source and Limit |
| --- | --- | --- |
| llms.txt gets you recommended | Not shown. Google says Google Search does not use it, and Ahrefs found 97% of llms.txt files got zero requests in May 2026 | [Ahrefs](https://ahrefs.com/blog/llmstxt-study/), 137,210 domains |
| Schema markup earns recommendations | Not shown for ChatGPT. Google says no special schema is needed for its generative AI features | [Google's guide](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), Google only |
| A GEO checklist lifts discovery | Not shown in one preprint: startups with high GEO scores were no more likely to show up | [Sharma, arXiv, Jan 2026](https://arxiv.org/abs/2601.00912) |
| Reviews and community posts matter | They are cited. In my 450 recorded answers of one software category, reddit.com was cited in 73, YouTube in 48, LinkedIn in 30, G2 in 16 and Capterra in 7. Citing is not proof it caused a recommendation | A GEO agency's 405-answer test, plus 45 brand-name answers |
| More backlinks mean more recommendations | Not established. Brand mentions correlated 0.664 and backlinks 0.218 with AI Overview mentions | [Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/), 75,000 brands, correlation only |
| A vendor can guarantee a recommendation | No documented basis. OpenAI gives no ranking formula | OpenAI's documentation |

### Two More Ways to Waste Money

The first is trusting a visibility score without the answers behind it. In one preprint, startups with high GEO scores were no more likely to show up, so ask any tool or agency for the exact questions, apps and recorded answers.

The second is publishing dozens of generic AI-written pages. That is my opinion, not a measured result: pages that answer one real buyer question are the ones worth the time.

Here is what that looks like. On 15 Nov 2026 a founder posted that Google's core update hit his site after he published one AI-written article a day on a domain with a rating under 20.

He is now keeping his best 30 pages and rewriting them by hand. That is Google, not ChatGPT, but it shows the cost of volume.

### Four Explanations That Are Usually Wrong

"They paid to be there" is covered by OpenAI's ad policy in the FAQ above.

"The model is biased" does not fit what the answers said: all 11 AI answers I collected described an evidence gap, not a verdict on quality.

"We need more content" is weaker than it sounds: in Seer Interactive's study of 4,124 cited pages from March to June 2026, 72% had been updated in the last year, but only 42% had been published in the last year, so refreshing a page you already have counts for more than publishing a new one.

"Our authority is too low" is not final either, because the founder with the very low domain rating reported about 50 users a month from ChatGPT.

### What This Means for Your Priorities

If you have a limited budget, do the free, documented fix first. Then do the work that makes you easier to understand and easier to verify: one plain description, pages for the exact questions buyers ask, and honest listings on G2, Capterra and the roundup articles the answers already cite.

Treat llms.txt and schema as optional housekeeping.

**Do this now:** if you only have an hour, do the first three fixes in the table above and skip llms.txt and schema until they are done.

## How Do I Know Whether Anything Improved?

Ask the same questions the same way before and after, then compare separate counts. A single score hides which one moved.

| Count | What It Measures | Why Keep It Separate |
| --- | --- | --- |
| Mentioned | Answers that contain your brand | Shows whether ChatGPT has you in view at all |
| Recommended | Answers that point the buyer to you | The result that matters most |
| Cited | Answers that use your page as a source | Shows your content is read even when you are not mentioned |
| Rivals recommended | How often each rival is recommended | Shows who is taking your place |
| Described correctly | Whether the answer gets your product, price and audience right | Wrong facts cost deals even when you are mentioned |

Write your baseline with the same four counts: mentioned, recommended, cited, and answers that recommended no one. The re-check uses the same five questions, the same three apps and five runs each.

### How to Tell a Real Change From a Swing

Compare ranges, not single numbers. If the before and after ranges overlap, the honest result is "no clear change yet".

Write down any model update or site change in the same window. And do not report citations as traffic: Bing's own documentation says a citation does not represent clicks or visits.

### How to Count Leads From AI

A count shows how often a rival is recommended. It cannot show a deal that never reached you. So ask the buyer.

My free check form has an optional "How did you hear about us?" field, and I read every answer.

Add the same question to your demo request form and sign-up form, with ChatGPT, Perplexity, Gemini and Claude as choices, and write down every answer.

### How to See ChatGPT Visits in GA4

When someone clicks a link inside a ChatGPT answer, ChatGPT adds a tag to the address.

When I opened Exibran from ChatGPT, the address ended in `exibran.com/?utm_source=chatgpt.com`.

OpenAI's help page says ChatGPT adds this tag to referral links so you can track them. In GA4, open Reports, then Acquisition, then Traffic acquisition.

Change the first column to Session source / medium and search for "chatgpt". My [AI Pick Rate guide](/blog/ai-pick-rate) shows each click.

This counts clicks only. An answer that recommends your rival and sends nobody to your site will not appear here.

## What Should I Tell My Boss or Investor When ChatGPT Recommends a Competitor?

Tell them three things: what a counted check shows, which fixes you are doing first, and that nobody can promise a placement.

1. **What we measured.** "On [date], I asked three AI search our [number] buyer questions [number] times each. That is [total number] answers. We were mentioned in [x] and recommended in [y]. The most recommended rival was recommended in [z]."
1. **What it means.** "This does not say the product is worse. AI answers change run to run, so I report the count and its range, not one screenshot."
1. **What we do first.** "First we make sure ChatGPT's crawler can read the site. Then we put one plain description and the price on every page. Then we get onto the pages the answers already cite. We re-run the same [total number]  on [date]."

Do not promise a placement, do not show one screenshot as proof, and do not say "it's random".

Say "we were mentioned in [x] of 75 answers, and the range says [low] to [high] percent". Bring the filled sheet and the list of cited pages. A counted sheet answers "how do you know?" before it is asked.

## How Long Does It Take?

No published study gives a reliable timeline for becoming a stable ChatGPT recommendation. Here is what I can show:

- **Crawler changes:** about 24 hours, per OpenAI's crawler documentation.
- **Everything else:** Tom Han went from not being mentioned in July 2026 to steady users from ChatGPT by November, about four months. It is self-reported and one case.

Treat anyone who gives you a number of weeks without a measured example with caution. Re-count on a fixed day each month, with the same questions and the same apps, and write down any site change in the same window.

## Who Can Do This, and Which Should I Choose?

You can do it yourself, use a tracking tool, or ask me. Choose by your situation, not by who is writing this guide.

| Option | Choose It If | Avoid It If | Cost |
| --- | --- | --- | --- |
| Do it yourself with the sheet above | One person can edit your site and run 75 answers | Nobody on your team can edit pages | Free |
| A tracking tool | You want weekly counts after you have fixes in place | You have no fixes yet, because a score with no action list does not change the answers | Otterly.AI from $29 a month for 15 prompts, Peec AI from about $95, Profound from $99 on annual billing for ChatGPT only ([published prices, Sept 2026](https://www.ud.hk/en/insight/article/ai-visibility-tool-pricing-2026-09-03)) |
| Exibran free check | You want the 75 recorded answers and the biggest gap before spending anything | You want a guaranteed placement | Free, within 48 hours |
| Exibran TRACED Diagnostic | You want a full report with fixes written for your pages | You only need Google rankings | $1,500, half to start and half when all five deliverables are delivered |

For a team with one person who can edit pages, I would run the sheet first, then ask for the free check if you want it counted for you. Exibran is not the right pick if you want a promised ranking or only care about Google traffic.

## Can Exibran Check Which Competitors ChatGPT Recommends Instead of My Company?

Yes. Exibran's free check asks ChatGPT, Gemini and Perplexity five buyer questions about your company, five times each, and sends you the 75 recorded answers by email within 48 hours.

### What You Send

Your email, product or brand name, and website. You can also add where your buyers are, your main competitors, and the questions your buyers ask AI. It takes a minute.

### What You Get in 48 Hours

- The 75 recorded answers, readable word for word.
- How often you were mentioned, recommended and cited, with the range.
- Which rivals were recommended instead, and how often.
- The pages the answers relied on, so you can see where you are missing.
- The biggest gap I see, written plainly, and two fixes you can apply today.

![Free Exibran Visibility Report](../../assets/images/posts/why-chatgpt-recommends-my-competitors/visibility-report.png)

### What the Free Check Does Not Do

It does not fix anything, it does not promise a recommendation, it does not review your Google rankings, and it does not require a call. Skip it if you only want Google traffic, a guaranteed placement, or a local-business listing check.

### What Happens Next, If You Want More

The TRACED Diagnostic is a fixed $1,500: half to start, and the other half only when all five deliverables are delivered. They are how often AI mentions you, why AI picks others, what to fix and what to write in order, technical fixes for your developer, and how to track it. The 90-day TRACED work is billed monthly with a 90 day minimum, and I give the price on a call. The free check never obliges you to either.

## Frequently asked questions

### Are My Competitors Paying to Be Recommended?

No. Ads now exist in ChatGPT, but OpenAI says ads do not influence the answers, that advertisers cannot rank or change responses, and that ads are labeled and shown separately below the answer. Paying gets a competitor an ad, not a recommendation inside the answer. What can help a competitor is being mentioned on other sites, which is the honest route in this guide.

### Does ChatGPT Use G2, Capterra, and Reddit?

It cites them, and in my data Reddit and YouTube were cited more than G2. Across 450 recorded answers for one software category, reddit.com appeared in the sources of 73 answers, YouTube in 48, LinkedIn in 30, G2 in 16 and Capterra in 7. A citation shows a page was used as a source. It does not prove it caused the recommendation.

### Can a Newer Company Be Recommended?

Yes. One founder who built his tool in June 2026 reported about 50 users a month from ChatGPT by November, on a site with a very low domain rating. It is self-reported, and it took link swaps, a few paid mentions and a set of new pages, so it is not a guarantee.

### Is This Just SEO?

Partly. In one study, 58% of Google's top-10 pages were also cited by ChatGPT, so a good rank helps and does not settle it. Google says optimizing for its own AI features is still SEO. ChatGPT adds two things: its crawler has to be let in, and other sites have to mention you.

### Does Ranking Number One on Google Guarantee ChatGPT Will Recommend Me?

No. Position Digital found that 58% of pages in Google's top 10 were also cited by ChatGPT, across 278 prompts, so a good rank helps but many top pages were not cited. Google and ChatGPT use different systems, so count your ChatGPT answers separately.

### How Do I Get ChatGPT to Recommend My Business?

Nobody can force it. Do three things in order: let ChatGPT's search crawler in, write one plain description with your price and use it everywhere, and get onto the pages the answers already cite. Then ask the same questions again and count the change.

### How Can I Check Whether ChatGPT Knows My Company?

Type your company name in a new chat and see how it describes you. Then ask five questions without your name. Knowing your name and recommending you are different, and in one check every answer that included the company name described it well while none of the 96 name-free answers mentioned it.

### Does llms.txt Help ChatGPT Recommend My Website?

There is no evidence that it does. Ahrefs found 97% of llms.txt files got zero requests across 137,210 domains in May 2026, and Google says its Search does not use the file. Letting OAI-SearchBot in is the documented step.

### Why Does ChatGPT Recommend Different Companies When I Repeat a Question?

Each answer is built fresh. AirOps found only 30% of brands stayed visible in back-to-back responses on average. Ask each question five times and count, instead of trusting one answer.

### Can an SEO Agency Improve My AI Visibility?

It can fix access, descriptions and pages, and show you counted answers before and after. It cannot promise a recommendation, because ChatGPT is an independent platform. Be careful of anyone who guarantees one.

### Does GA4 Track ChatGPT Traffic Automatically?

Mostly. ChatGPT adds utm_source=chatgpt.com to links inside its answers, so GA4 shows chatgpt.com as a session source when someone clicks. It cannot show an answer that recommended a rival and sent no one to you, so ask buyers how they found you.

### Does Bing Matter for ChatGPT Recommendations?

Less than it used to. Peec AI's September 2026 study found ChatGPT now runs its own search index, with Bing showing up mainly in Deep Research. Letting OAI-SearchBot in is still the documented step.
