All posts

What to Do When ChatGPT Recommends Your Competitor Instead of You

Cover illustration for “What to Do When ChatGPT Recommends Your Competitor Instead of You”

When ChatGPT recommends your competitor instead of you, the answer is not to keep asking until your name appears. How models learn about brands and retrieve evidence makes this a structural problem. The response needs the same discipline. Measure where competitors displace you, identify the content and citation gaps behind those results, and track whether targeted changes improve your share of voice over time.

Why ChatGPT recommending your competitor instead of you is a structural problem, not a fluke

You ask ChatGPT a question relevant to your category. Seeing your competitor recommended instead of your brand can feel like a mistake, especially if your business is well established. The model draws on patterns learned during training, supplemented by web retrieval when enabled. A stronger representation of your competitor gives it more reason to recommend that competitor.

The loss is difficult to see in ordinary analytics. There is no missed impression or failed click recorded on your site when an assistant leaves you out of its answer. Unless someone watches the recommendations themselves, no one sees that the brand has been left out. Even then, a single result can mislead because model answers change between runs. A favorable answer does not establish a secure position, just as one omission does not establish the size of the problem.

How AI models actually decide which brands to surface in a recommendation

ChatGPT, Gemini, and Claude take different paths to evidence. ChatGPT draws on sources including Wikipedia, major publications, Reddit, and community and review platforms. It gives authoritative list mentions the greatest weight, followed by awards and affiliations, online reviews, customer examples, and social sentiment. Gemini starts with sites that rank well in Google Search for the category, then adds reviews, case studies, and product data. Strong Google Business profiles and appearances in “best of” lists therefore help brands surface there.

Claude has a different retrieval path again. It uses Brave Search as one of its web search providers, together with TurboPuffer and Parallel Web Systems, making presence in Brave’s index relevant to retrieval through that route. It gives particular weight to authoritative list mentions and accreditations, and is less likely to name a single “best” brand. As a result, a brand can appear in most ChatGPT answers for a query and never appear in Claude’s answers to the same question.

Because the models differ, no single fix works for all of them. They also make it necessary to distinguish three outcomes that often get treated as one. A mention means the brand name appears, while a citation means a specific piece of content is referenced as a source. Correct framing means the model describes the brand accurately and favorably. Tracking the name alone leaves both the evidence behind the answer and the quality of the recommendation out of view.

What AI share of voice measures and why a one-off check cannot show it

AI share of voice is the number of brand citations divided by the total number of category citations across a defined prompt set. Relative position matters because category leaders receive a disproportionate share of AI mentions, leaving the average brand with much less visibility. Seeing a brand somewhere in an answer does not show whether it is gaining ground against those leaders. Nor does it show whether the assistant presents it as a serious choice.

Brand24’s October 2026 roundup notes that AI referral traffic converts at a meaningfully higher rate than non-AI traffic. The framing of a recommendation therefore has commercial consequences. A frequently mentioned brand can still lose if the assistant describes it unfavorably and presents a competitor as the clear choice. Negative visibility can be worse than absence.

To measure that position, track each engine separately and repeat the same prompts over time. Identical questions can return different brand lists, so a single answer is only one sample of the model’s behavior. It cannot establish share of voice or reveal a trend. Cross-engine averages can also hide a strong result on one platform alongside a serious gap on another. Frequency, relative citation share, and sentiment need to remain distinct parts of the picture.

Building a prompt library that actually surfaces where the displacement is happening

Start with questions buyers would ask if they had never heard of your brand. Putting your own name into discovery prompts artificially inflates visibility, because the question itself supplies the brand you are trying to test. Category discovery prompts should reveal which businesses the model names unprompted. Problem-solution prompts should describe a need without naming the category directly. These test whether the assistant connects that need to your solution at all.

Comparison prompts serve a different diagnostic purpose. Ask the model to compare a competitor with another named rival, then see whether your brand enters the competitive set. Keep those results apart from unprompted discovery results, since the questions test how the model behaves in different cases. A model may know your brand when asked directly but fail to introduce it when a buyer is considering alternatives. That gap is more useful than a flattering answer to a question that already names you.

Coverage should also follow the buyer’s journey. Include discovery, consideration and comparison, evaluation, and switching prompts, because each stage can expose a different omission. Run each prompt several times per session so that one answer that includes or leaves out the brand carries less weight in the result. The library is the foundation of the monitoring program. Automation can repeat and organize these tests, but it cannot make a poorly scoped set of questions representative of your buyers.

How systematic monitoring tools turn prompt results into trend data you can act on

Entering prompts by hand can help with a first look. It is not a monitoring strategy. Opening ChatGPT, entering a question, and recording the answer does not provide the volume, frequency, and consistency needed for meaningful trend data. Monitoring tools automate schedules across models and aggregate results into mention rate, share of voice, sentiment, and competitor comparisons. They turn a pile of answers into a record that can be compared over time.

That record needs more than a brand count. Track whether the brand appears, whether it is the first recommendation or an afterthought, how it is described, and which sources support the answer. Yotpo’s November 2026 roundup notes that some tools query model APIs while others capture answers from consumer apps. Those are not interchangeable views, because consumer apps can include live search results and citations that API responses sometimes omit.

Lettertrace is a free, open-source, permissively licensed option built for this workflow. It is self-hostable and uses a bring-your-own-key approach, with no usage markup or vendor lock-in. It automates prompt variation and queries ChatGPT, Claude, Gemini, Google AI Overviews, and Perplexity Sonar. Results are aggregated into visibility, prominence, sentiment, and share of voice metrics by topic and model over time. With Supabase storage that the user owns, the monitoring data stays in the team’s own database.

Tool selection should follow the reporting job. Compare which engines come with the base tier and which need add-ons, whether tools save raw responses for auditing and track competitors, and whether data can be exported or stays in vendor systems. The Brand24 and Yotpo roundups describe a rapidly developing category that spans entry-level trackers and enterprise suites. The most feature-rich option is not automatically the right one. The useful choice covers the engines your team needs and preserves the evidence behind its reports.

Reading the competitor benchmarking data to identify the specific gaps driving displacement

A competitor appearing consistently across informational and commercial prompts is a diagnostic signal. Covering more topics may give the model more useful material to retrieve when it answers related questions. Authoritative third-party citations may make the competitor better represented in sources the model trusts. Consistent descriptions across those sources may also give the model a clearer picture of what the competitor does.

Citation analysis helps distinguish those possible causes. When an answer cites a competitor, inspect the analyst list, review platform, comparison article, or community discussion behind it. The Banks AI Visibility Index illustrates why this is worth doing. Across consumer-intent prompts, bank-owned domains accounted for a small share of banking citations, while Bankrate, Investopedia, and Wikipedia accounted for the large majority. In that category, looking only at banks’ own websites would miss most of the cited evidence.

A competitor with fewer mentions may still receive more favorable or specific descriptions, giving buyers a clearer reason to choose it. That is a different gap from simple absence. Benchmarking should therefore connect each weak result to its topic, engine, cited sources, and framing. The output is not a verdict that your brand is generally weak, but a diagnosis of where its supporting evidence falls short.

The content and citation changes that shift AI share of voice over time

Begin with a consistent description of your brand. If your website, LinkedIn, Crunchbase, and third-party bios describe the business differently, they provide weaker combined evidence of what it is and whom it serves. Aligning those descriptions is the lowest-cost starting point because it requires discipline without a large technical project. It gives models a clearer identity to recognize across sources. This addresses an identifiable weakness in how the brand is represented.

If models repeatedly cite particular review platforms, analyst lists, comparison articles, or community threads, the goal is to earn accurate, favorable coverage in those sources. Trying to get more mentions anywhere can miss the places that already shape recommendations. The banking example shows why this work cannot be confined to your own domain.

Include relevant statistics, quotations from credible sources, and explicit citations, and keep the material current, since AI systems tend to cite fresher content than traditional organic results. Structure comparisons and other suitable pages so that their information is easy to extract, including ordered or unordered lists where appropriate. AI Overviews and similar surfaces disproportionately draw from structured content. Comprehensive, interlinked topic coverage also creates more opportunities for retrieval on adjacent questions than a narrow set of pages does.

Address robots.txt configuration, an llms.txt file, and clean indexation so that access is not the barrier. Judge those changes through monitoring over time, keeping expectations of a quick jump after an edit in check.

Turning monitoring into a feedback loop that catches sentiment shifts before they compound

A brand may appear more often while the model’s description becomes outdated or unfavorable. Old funding announcements, early-stage reviews, or out-of-date product descriptions can make a mature product sound early-stage. A visibility report alone would miss that deterioration. Sentiment and source tracking show whether the assistant’s picture of the business is keeping pace with the business itself.

That check also needs an engine-specific view. Claude tends toward cautious, balanced descriptions, while ChatGPT is more assertive in its recommendations. A neutral description from Claude and a negative one from ChatGPT point to a ChatGPT-specific problem that an average could conceal. Scheduled monitoring establishes the baseline and reveals changes, including a new competitor entering recommendations or a negative review source becoming a primary citation.

After targeted content and citation changes, examine share of voice and sentiment on the prompts those changes were meant to affect. Measurable improvement gives the team evidence for its next decision and something useful to report upward. Then the cycle starts again. AI brand visibility now demands the sustained rigor SEO teams applied to search rankings, with a habit of measuring, diagnosing, changing, and measuring again that occasional manual checks cannot match.