Published Updated 12 min read
OpenAI model via Codex CLI with web search, not the ChatGPT app, 01-02.11.2026, 13 software questions.
Why does ChatGPT recommend your competitor and not you? The cause can often be found in a few minutes. Here are 5 common reasons, the same for Perplexity and Google AI, and a check for each.
Card needed. Cancel before day 7 and nothing is charged.
Your company (Your company)Competitor
Is Kelderwick Cycles a good place for an e-bike service in Bristol?
Yes, Kelderwick Cycles (Your company) services most e-bike brands.
You could also try Gorsefield Bike Works and Temple Spoke Co.
ChatGPT names a competitor when it can find and read clear facts about them for the customer's question.
Rivals by default Asked about one product by name, AI still added rivals. In our runs: 10 of 10 times for a video-call tool, 7 of 10 for a team chat tool.
One answer is a sample Ask 3 times in new chats before you decide anything.
Checks you can do today Each reason below comes with a check that takes a few minutes.
AI search looks for a page that fits what the customer typed. If your rival has that page and you do not, AI is more likely to quote the rival.
Customers type long questions. They add their size, their city, their budget or their problem.
A general home page rarely answers "Which bike shop in Bristol services e-bikes on a Saturday?". A rival's page called "E-bike service, open Saturdays" answers it in one line.
For the kinds of pages AI quoted in our tests, see the guide to pages AI cites.
Which bike shop in Bristol services e-bikes on a Saturday?
kelderwickcycles.example
Your site: no matching page
gorsefieldbikes.example
/e-bike-service
E-bike service, open SaturdaysCited
gorsefieldbikes.example/e-bike-service
Scheme: the 3 parts of the question match one page on the rival's site and no page on yours.
Use the words customers use on the phone or in e-mails, not your own product names.
For each question, open the one page that should answer it. Read only the first two lines.
No page, or only a vague one? That question is the most likely place your rival wins.
If your robots.txt or firewall blocks the bots that AI search uses, AI cannot read your pages. It then names companies it can read.
Each AI engine sends its own bot.
| Engine | Its bot | What the bot does |
|---|---|---|
| ChatGPT | OAI-SearchBot | ChatGPT search uses OAI-SearchBot to read pages for its answers. |
| Perplexity | PerplexityBot | Perplexity uses PerplexityBot. |
| Google AI | Googlebot | Google AI Overviews use pages from normal Google Search, so Googlebot must be allowed. |
| OpenAI training | GPTBot | GPTBot is a different OpenAI bot: it collects text for training, not for search answers. |
User-agent: OAI-SearchBotDisallow: /User-agent: PerplexityBotAllow: /User-agent: GooglebotAllow: /
Scheme: OAI-SearchBot stops at the gate, PerplexityBot and Googlebot reach the site.
Go to yoursite.com/robots.txt. Look for "Disallow: /" under OAI-SearchBot, PerplexityBot, Googlebot or "User-agent: *".
Ask your host or developer one question: do we block AI bots or unknown bots?
Open a key page with JavaScript switched off in your browser. Text that disappears may be missed by bots too.
Want to see who AI names for your own questions? Nenn checks it for you every week. Start 7-day trial
AI tends to name companies it already sees on many websites. If reviews, lists and local news mention your rival and skip you, the rival comes first.
72 of 75list runs where the first search already held a brand from the answer
We looked at the model's first web search in 75 list runs. In 72, the search words already held a brand name that later appeared in the answer.
So the shortlist often seems to exist before any page is read.
0.664how closely mentions went with AI Overview visibility (Ahrefs)
0.218how closely backlinks went with it, same study
The Ahrefs figures are their study, and they show a link, not a cause.
In May 2025, Ahrefs studied 75,000 brands in Google AI Overviews.
Web mentions of a brand went together with visibility more than backlinks did: 0.664 against 0.218.
That is their study. It shows a link, not proof that mentions cause the answer.
Scheme: 4 websites mention the rival, 1 of them mentions you.
Search your company name in Google and add -site:yoursite.com. Count the pages that are not yours.
In an answer that names your rival, open the links next to the name. Note which websites they are.
Missing from those sites? Ask real customers for honest reviews there. Never pay for reviews.
AI answers change from run to run. A rival named once may be gone next time, and you may show up instead.
9different orders of the same top 5
We asked one list question 10 times and got the top 5 in 9 different orders.
Answers also differ by engine, by country and by the person's chat history.
So one screenshot from a friend is a sample of one.
Before you rebuild your website, read why one look at AI is not enough.
Run 1
Run 2
Run 3
Ask the same question in 3 new chats. Log out first if you can.
Repeat on ChatGPT, Perplexity and Google. Each engine searches in its own way.
A rival named in 3 of 3 answers is a pattern. Named in 1 of 3 is too early to act on.
Being named in the question does not keep rivals out of the answer.
We asked whether a video-call tool was a good choice for a company with 20 people.
10 of 10answers opened with yes and still named two rival video-call tools
7 of 10answers named rivals, after 10 of 10 opened with yes
| What we asked | Runs | Opened with yes | Named rival products |
|---|---|---|---|
| Is this video-call tool a good choice for 20 people? | 10 | 10 | 10 |
| Is this team chat tool a good choice for 20 people? | 10 | 10 | 7 |
OpenAI model via Codex CLI with web search, not the ChatGPT app, 01-02.11.2026, 13 software questions. Each run in a new chat. Product names removed; we show the product category.
A rival next to you is normal. What matters is what AI says about you, and which page of yours it quotes.
All 10 answers opened with yes. All 10 also named two rival video-call tools. For a team chat tool, all 10 answers opened with yes, and 7 named rivals.
How we got these numbers: on 1 and 2 November 2026 we put 13 software questions to an OpenAI model that can search the web. 115 runs in total, each in a new chat. This is not the ChatGPT app, so your own chats can differ. Product names are replaced by their category.
Nenn (nenn.io) is a SaaS from Hamburg. Once a week, Nenn asks three AI engines the same questions your customers ask. For each one it shows which companies get named.
For each question you see the full answer, the names in it and the pages it cited.
Visibility
Question: Which bike shop in Bristol services e-bikes on a Saturday?
A rival's page cited for the exact question points to reason 1.
Directories that list your rival and not you point to reason 4.
Sources
6 cited domains · 3 engines · Last 7 days
Nenn lists the fixes for your questions, biggest gap first.
For reasons 2 and 3, its website check tests AI bots, the firewall, text without JavaScript, prices as text and your name and address.
For a missing page, Nenn drafts it. You read it and press Approve.
Nenn suggestions
RefreshSee all suggestions
Saturday hours right in AI answers
0 of 32 of 3after the next check
E-bike service on a Saturday
Kelderwick Cycles services e-bikes every Saturday from 9 to 4. We work on most e-bike brands.
Do you service all e-bike brands?
We service most brands. Call us if you are not sure about yours.
Nenn screens with invented Bristol bike shops.
See competitor analysisYes. Every check on this page works by hand. The hard part is repeating it every week, on three engines, from your customers' country.
| Task | By hand | With Nenn |
|---|---|---|
| Ask your questions on 3 engines | About 30 chats for 10 questions, typed one by one | Asked for you every week |
| Keep names and links | Copied into a sheet | Saved with the full answer |
| Spot the reason | Open every cited page yourself | Rivals and cited pages listed per question |
| Fix it | Write the page yourself | Draft to approve, then copy it or publish it (publishing: Growth and up) |
| See what changed | Repeat everything next week | The next weekly check shows it |
For a full manual method with a log sheet, use the step-by-step AI check.
Common causes: AI finds no page of yours that fits the question, or it cannot read your site. For ChatGPT, check that OAI-SearchBot is allowed in your robots.txt. A young company with few mentions on other sites is also easy to skip.
They can, as one kind of mention on other sites. Still, only 11 of 366 citations in our runs went to review or list sites. So reviews work best next to clear pages of your own, not in place of them.
Not always, but well-known names start ahead. In 72 of 75 list runs, the model's first search words already held a brand name from the final answer. Narrow questions, like one city or one service, have fewer good pages to pick from.
Each engine runs its own searches and picks its own sources. One may find your page, the other a directory that lists only your rival. Check each engine on its own, and open the links each one shows.
Open the links next to the wrong claim to find the source. Fix the fact on your own site first, then on your profiles and in directories. If another site is wrong, ask its owner for a correction. In Nenn, fixing wrong claims is one of the four task channels.
It can help AI find you for "this or that" questions. Keep every claim true and checkable. In Germany and the EU, comparative advertising has strict rules, so ask a lawyer before you name a rival.
Try Nenn on your own customer questions for 7 days. You see which rivals AI picks, the pages behind each answer and the first thing to fix.
Card needed. Cancel before day 7 and nothing is charged.
Which bike shop in Bristol services e-bikes on a Saturday?
Three next steps, from the other article to the tools that run the checks for you.
Six steps to become the page AI finds and quotes.
Blog post
Run the same question many times and see how the names move.
Tool
See who AI names for your questions, and what their pages have that yours lack.
Feature