Why AI Recommends Your Competitor Instead of You and How to Flip It

Why does AI recommend our competitor instead of us?

An assistant tends to name the brands it holds the most public evidence for, and your competitor likely has more of it. In our own runs, ChatGPT's searches already named vendors, while Claude searched the category (see How we tested).

Why does AI already know our competitor but not us?

Before any search runs, a model already ties certain brands to your category. Those ties come from the data it was trained on, including public web text, up to its knowledge cutoff.

OpenAI says its foundation models are built from public internet information, partner data, and data from users and trainers. The public part can include webpages, forums, blogs and posts. Its GPTBot crawls content that may be used for training, and Anthropic says ClaudeBot collects web content that could contribute to training.

Cutoffs differ. OpenAI lists GPT-5's knowledge cutoff as Sep 30, 2024, and Anthropic lists Jun 2026 for its current Claude models. A brand that grew after a cutoff is likely thin in that model's memory, however good its site is today.

Memory can shape shopping answers even when search runs. OpenAI's shopping help page says that when choosing which products to show, ChatGPT considers model responses generated before it looks at any new search results.

What does an AI engine search for when a buyer asks?

When an engine searches, it turns the buyer's question into its own queries, reads what comes back and cites some of those pages. The buyer doesn't write those queries, yet they decide which pages get read.

• ChatGPT. OpenAI's help center says ChatGPT may search when a question would benefit from current information, and, when it uses search partners, typically rewrites it into one or more targeted queries. Answers that use search may include citations.
• Claude. Anthropic's documentation says Claude decides when to search, can search repeatedly in a request and cites its sources. Questions about specific organizations or products that might have changed trigger searches, and comparative research can use 10 or more searches.
• Google. Google says AI Overviews and AI Mode may fan a question out into related searches, retrieved through core Search ranking systems.

Ahrefs notes that because of fan-out, a cited page may not rank for the buyer's original prompt.

Why does AI cite our page but recommend someone else?

Reading your page and recommending your brand are two different decisions. An engine can take facts from your page and still name a rival that appears on it.

Ahrefs tested this with 34 self-promotional pages and 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot. When AI cited a page promoting a conference, it still skipped that conference 43% of the time and recommended a competing event from the same page.

In the same experiment, a new brand and an established one drew their new mentions from different places. For the new brand, 82% of new mentions appeared in answers citing Ahrefs' own pages; for the established brand, only 6% did, and 94% came from third-party content.

Ahrefs concludes that self-promotional content seems to help most when the brand is already a believable answer. The new brand appeared in 66.4% of answers to an on-brand query and 15.8% for a broader one. Count only citations and a losing position can look healthy.

Do AI engines favor the bigger, better-known brand?

In one study, yes, but mainly as a tiebreaker when nothing else separated the products.

The study, Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems, set a real brand against fictional ones on GPT-4o-mini, Claude Sonnet and Gemini 3 Flash.

• Identical specs, known brand wins. Well-known brands were recommended 100% of the time when specifications matched, and across 670 valid trials no fictional brand was picked.
• A small edge breaks it. The dominance disappeared with less than a +0.1-star rating advantage for a competitor. The fictional brand won half the time with a 1.6x review count or a 7.3% price discount.
• Specs explain most of the ranking. Product parameters explained 82.4% of ranking variance and brand identity 1.2%.
• Brand is a tiebreaker. The authors say it counts most when the model has no other signal, and that the real barrier for new brands is a lack of distinguishing information.

The products were consumer goods: skincare, then USB-C cables and AA batteries across 3,840 calls. Software has fewer star ratings and longer feature lists, so for business software the result is only an indicator.

The authors' advice to established brands is better product content; for a challenger, the same logic says to publish what makes you different.

What makes AI pick one similar product over another?

Evidence about each product decides it, since matching feature lists cancel out. The table lists what an engine can check and how to see where you stand on each.

Which websites does AI use to decide which brands to recommend?

Pages other people publish about your category have strong ties to who gets named: branded mentions, community threads and best-of lists. Backlinks and page counts track it far less.

Brand mentions track AI visibility more closely than links

Ahrefs analyzed 75,000 brands across ChatGPT, AI Mode and AI Overviews. Branded web mentions correlated with AI visibility at 0.66 to 0.71, while link metrics showed very weak correlations.

A site's page count had almost no relationship, at about 0.194, in the same Ahrefs study.

In an earlier Ahrefs study of AI Overviews, brands in the top 25% for web mentions averaged 169 AI Overview mentions, against 14 for the next quartile. We cover how to earn that coverage in the earned media sources AI cites.

Reddit and community threads

In Ahrefs' September 2026 US data, Reddit was the most-cited domain in ChatGPT at 16.8%, Perplexity at 21.6% and Gemini at 28.5%. In Copilot, Amazon led instead.

Reddit reaches the models through deals too. Google's February 2024 agreement lets it display and train on Reddit content, and OpenAI agreed in May 2024 to bring Reddit content to ChatGPT. Even GPT-2's WebText dataset came from Reddit links with at least 3 karma.

Take part as yourself: answer questions in the threads your buyers read and say who you work for. Fake accounts and planted praise can get removed, and buyers notice.

Best-of lists and roundups

Ahrefs, citing Glen Allsopp's analysis of 750 ChatGPT prompts, reports that "best [category]" listicles were the most cited source type.

Its own research linked ranking high on third-party lists with appearing in ChatGPT responses. Our guide to getting a SaaS product into AI best-tools lists covers how to earn an accurate place.

Is gated content keeping us out of AI answers?

It can. Pages behind a login or form often can't be crawled, so the material that defines your category may never reach the engines.

Google explains that some pages go uncrawled because they need a login, and its AI guide says its generative models use publicly accessible, crawlable content. Google's advice is to allow crawling in robots.txt and in any CDN, and to keep important content in text form.

Bot rules can shut you out too. OpenAI says sites opted out of OAI-SearchBot are not shown in ChatGPT search answers. Anthropic says blocking Claude-SearchBot may reduce visibility in its results, and blocking Claude-User stops Claude fetching your pages for a user's question.

• Publish category-defining material as HTML. Methodology, benchmark findings, definitions and evaluation criteria belong on public, indexable pages.
• Keep forms for bottom-funnel assets. Calculators, implementation templates and proposals can stay gated where the exchange is fair.
• Move facts out of images and PDFs. Pricing tiers inside a graphic are harder to quote than the same facts in text.

How do we get AI to place us in the right category?

Describe the company and its category in the same words everywhere a model reads about you: your site, profiles, directories, review sites and press. One description lets an engine file you; mixed ones leave it unsure.

Google says Organization structured data on a home page can help it disambiguate an organization, and sameAs can point to your profiles on social or review sites.

Google also warns that seeking inauthentic mentions is not as helpful as it might seem, so consistency has to come from real profiles and coverage. Our guide to building a brand entity AI engines recognize walks through the audit.

How do we get AI to recommend us over the market leader?

Start with questions where the leader's fame helps you, then own one segment outright. A generic "best" question is the hardest place to unseat an incumbent.

Win "alternatives to" questions first

A request for alternatives to the market leader gets a list of substitutes built around the leader's name. A slot on that list is usually easier to win than the leader's place.

Publish a comparison page that is honest about fit, including the buyers who should pick the leader; see a comparison page AI engines will cite for a challenger.

Own one sub-category or segment

Choose a use case, vertical or company size the leader serves in general terms and you serve in detail, and become the plain answer there. That is the kind of distinguishing information the brand-bias study found new brands lack.

Why do ChatGPT, Gemini and Perplexity name different brands?

Their sources differ and shift from day to day. An April 2026 arXiv study found cited source sets overlapped by only 34 to 42% between consecutive days, with AI Mode concentrating citations most (Gini 0.782) and Perplexity least (0.671).

The study's authors say answers can vary across runs, prompts and time, and recommend at least 7 runs per prompt per day for brand monitoring. Our guide to measuring AI visibility sets that up. Google adds that AI Mode and AI Overviews may use different models, so their answers vary.

The brands overlap more than the sources do. Ahrefs found the assistants in its brand study largely mention the same brands, with an output overlap correlation of 0.779. It suggests AI Mode acts as a consensus engine favoring brands people already know, and says ChatGPT may be the best entry point for smaller brands.

How do we flip the recommendation to us?

Run the same sequence for each engine and each buyer question.

1. Diagnose per engine and question. Ask what your buyers type, over repeated runs, and record the brands named, the searches run and the pages cited.
2. Separate recommended from cited. See the citation section above.
3. Fix category clarity. Use one description everywhere (see the category section).
4. Make your main pages public and crawlable. Ungate category-defining material (see the gated-content section).
5. Earn mentions on the sources engines used. Start with the roundups, reviews and threads step 1 found.
6. Publish specific distinguishing facts. Use the near-tie table above. Google's guide says unique, non-commodity content will likely influence generative AI search presence in the long run more than any of its other suggestions.
7. Measure again. Re-run step 1 and compare with the first run.

Should you run this in-house or with an agency?

• Team time. Repeated runs and page fixes are ongoing work, so someone has to own them every month.
• Tooling. Asking several engines the same questions many times over needs software, or a partner that runs it for you.
• Outside placements. Reviews, roundups and community threads take relationships and patience a content team may not have spare.

If those gaps are real, you can see how we run AEO. As an AEO and GEO agency, OnlyAEO keeps separate counts for answers that recommend a brand and answers that only mention it in text or only link to it as a source.

Our Context Graph gives each client one record that every article, video and post is written from, covering its facts, buyers, competitors and the claims it can and cannot make.

FastTrackr AI started at #338 of 458 brands with 0.1% AI visibility. Roughly 100 days on, it stood #3 among its category's most-cited brands and #4 among the category's most-cited sources.

We went from being essentially invisible in our category to the #3 most-cited brand inside one quarter.

OnlyAEO offers no results guarantee and promises no specific rankings or citations. In most cases the first measurable results come after 60 days or more.

How we tested

We asked ChatGPT and Claude, with web search on, for the best customer onboarding software for a SaaS company selling to businesses, and the best project management tools for a marketing agency with a couple of hundred staff.

We run this engine check before writing any article: both assistants get the buyer's question, we read each page they cite in full, and each search they ran becomes a section.

• ChatGPT searched vendor by vendor. On both questions, every search it ran already named a vendor, and it then cited mostly those vendors' own product, pricing and help pages.
• Claude searched the category. It ran a single broad "best" search per question, cited roundup lists and named brands from them. It warned that many best-of lists are written by vendors who rank themselves first.
• Claude flagged what it could not source. On the agency question it said Adobe Workfront, a common enterprise choice, did not appear in its results, so it couldn't point to a source for it.

We didn't repeat the runs, so they only show what the assistants did on that day.

If these runs are typical, the two patterns need two fixes: for ChatGPT, a stronger public record over time; for Claude, a place on the roundups that rank for the category question.

More questions about AI recommending your competitor

Can we block AI training and still appear in ChatGPT search?

Yes. OpenAI treats search inclusion (OAI-SearchBot) and training (GPTBot) as independent settings, so a site can allow search while disallowing training. OpenAI says its search systems can take about 24 hours to adjust after a robots.txt update.

Can we pay to have ChatGPT recommend our products?

OpenAI says ChatGPT product results are selected independently and are not ads or influenced by its partnerships. Merchants are ranked on factors such as availability, price, quality and whether they are the maker or primary seller of the item.

Where do the review summaries in ChatGPT shopping answers come from?

OpenAI says ChatGPT's product review summaries are model-generated from reviews on public websites. Reviews your customers leave on public sites can feed what ChatGPT says buyers like and dislike about your product.

Will bolder marketing claims make AI prefer us?

In the incumbent-advantage study, authority-style language, including fabricated clinical claims, broke the known brand's hold at a value equal to +0.17 rating points. The authors limit their advice to real authority signals and vague authority language, and call made-up claims potential false advertising, and when every brand used the same tactic, individual payoff, as the authors measured it, fell from +0.802 to +0.007.

Can a tool show exactly why Google's AI chose our competitor?

Google says no third-party tool has access to its internal ranking or AI systems.