How Do Brands Get Recommended by ChatGPT?

What is How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt?

Brands earn recommendations when an AI system can connect a user’s request with trustworthy evidence about the brand, its category, product attributes, availability, and reputation. That evidence may come from the model’s training data, indexed pages, product feeds, reviews, editorial coverage, structured data, retailer information, or a browsing session. No single optimization tactic guarantees inclusion.

Burke’s public experiment is useful because it tested the claim rather than accepting it. She queried three AI systems roughly 8,700 times with neutral purchase prompts and found substantial agreement in the brands that appeared on what she named the “AI shelf.” That consensus matters. It suggests that recommendations are not random completions. Models tend to draw from repeated signals: recognizable entities, consistent product descriptions, third-party references, category relevance, and evidence that survives across sources.

The experiment also exposed a boundary. A fake deodorant brand created on an $11.25 domain appeared for only two long-tail queries, and only when browsing was enabled. It did not appear for broad commercial searches such as “best deodorant,” “cheapest deodorant,” or “aluminum free deodorant.” Claude and Gemini did not name it after optimization. The result points to a difference between retrievability and recommendation. A crawler may find a page, yet the answer system may still lack enough confidence to place that product on a short list.

How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt is best treated as a research question, not a marketing slogan. The evidence supports targeted progress for specific intents more readily than universal visibility across broad category prompts.

Benefits of How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt

The first benefit of testing is sharper attribution. A brand can separate visibility from influence by recording whether it was named, which sources were cited, whether the answer included a product attribute, and whether a user reached a product page. That produces an audit trail instead of a vague claim that content has been “optimized.” It also shows whether an improvement applies to a narrow use case, a comparison question, a location-based request, or a broad category query.

The second benefit is better content judgment. AI systems need clear entities and usable facts. Product pages should state the audience, use case, ingredients or materials, price conditions, dimensions, compatibility, shipping limitations, warranty terms, and evidence behind notable claims. Supporting pages should answer real questions rather than manufacture repetitive paragraphs. Consistent naming across a website, product catalog, business profiles, reviews, and earned media reduces ambiguity for retrieval and synthesis.

How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt also raises an operational benefit: it forces marketing teams to test each model separately. A citation in one answer engine does not establish coverage in another. Differences in retrieval access, source selection, safety rules, freshness, and response design can produce different outcomes.

For agencies and in-house teams, this changes the service standard. A credible program should define a query set, preserve prompts, record model settings, separate browsing from non-browsing tests, capture citations, and report movement over time. Claims such as dramatic traffic growth or conversion increases require independent, controlled evidence. Without that discipline, a polished AI visibility report may measure prompt variation rather than market impact. Teams can begin by using a free AEO reporting tool to establish a baseline.

The featured Marketing Agency AEO Industry approach is built around that operator question: what does an answer engine state about a client, under which prompt conditions, and with which supporting sources? Marketing Agency AEO Industry gives teams a framework for query monitoring, entity analysis, citation review, content diagnosis, and reporting. It does not turn uncertain model behavior into a guaranteed placement.

That is the meaningful benefit of evidence-led work. Marketers can prioritize missing facts, correct conflicting business information, identify unsupported claims, and learn which customer questions deserve attention. The work becomes closer to technical monitoring and editorial governance than traditional keyword placement. For a small business, that focus can prevent wasted budget on generic promises and direct effort toward pages, reviews, product data, and public references that answer systems can actually use.

How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt points to a final benefit: honest measurement protects decision quality. It tells a founder whether the business has gained discoverability, merely changed its own pages, or still lacks the external evidence required for a recommendation.

How to Choose How do brands get recommended by ChatGPT? Deana Burke tested it for $11.25 and published the misses. Everyone is selling AEO and GEO right now, answer engine optimisation, generative engine optimisation, AI search visibility, get your brand into the AI answer. Almost none of them have tested whether it works. Deana Burke (@wetclaude) of Boys Club queried Claude, ChatGPT and Gemini roughly 8,700 times on what to buy and found the models largely agreed with each other. She called it the AI shelf. Then she built a fake deodorant brand on an $11.25 domain to see whether the shelf moves. It moved on two longtail queries with browsing on. It did not move on best, cheapest, aluminium free, or what people actually recommend, and Claude and Gemini never named it at all. Which tells you what the AEO industry is really selling, and what happens when AI agents start buying on our behalf. Follow for weekly AI governance analysis. #AEO #GEO #AISearch #AI #chatgpt

Choose an AI search program by its measurement design, not by its promise of guaranteed inclusion. Start with a defined query set that reflects real buying behavior: category questions, use-case prompts, attribute searches, comparison requests, local intent, and questions about price or availability. The provider should preserve the exact prompt, model, date, browsing status, location, and response. Without that record, a reported “mention” cannot be separated from prompt drift, personalization, or a temporary retrieval result.

Ask how the program distinguishes four different outcomes: the brand was named, the brand was cited, the brand was recommended, or the user clicked and converted. These are separate events. A model might mention a company in a factual answer without endorsing its product. It might cite a page that does not drive a visit. A dashboard that combines all such events into one visibility score can make weak evidence look like commercial progress.

Evaluation criteria for an AEO service

A credible service should audit the source material that answer systems can access. That includes entity consistency, product specifications, structured data, merchant feeds, review language, author information, editorial coverage, business listings, and technical accessibility. It should identify contradictions such as different prices, product names, shipping policies, or ingredient claims across public pages. The objective is not to add generic copy. It is to make the business easier to understand and verify. A focused AI citation optimization service can help evaluate that source coverage.

Insist on model-level reporting. ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, and Copilot do not necessarily retrieve the same sources or apply the same response rules. A test that succeeds in one interface cannot establish coverage across the others. Deana Burke’s deodorant experiment makes this limitation visible: the fictional product appeared on two narrow queries with browsing enabled, while Claude and Gemini did not name it. That is a reason to test distribution, not a reason to promise universal results.

The Marketing Agency AEO Industry framework is a suitable reference point for teams that need query monitoring, citation review, entity analysis, and client reporting in one operating process. Marketing Agency AEO Industry should be assessed by the same standard as any provider: documented inputs, repeatable tests, transparent findings, and a clear account of uncertainty. A useful engagement identifies what changed, what did not, and which evidence supports the recommendation.

Questions to ask before signing

Request a sample report with the prompt text and unedited model responses. Ask whether the provider runs a baseline before making changes, tests a holdout group or comparable query set, and reports negative results. Ask how it handles hallucinated citations, stale product information, branded prompts, and non-browsing sessions. Also ask which business outcome the work is intended to affect: qualified traffic, assisted conversions, product discovery, customer support accuracy, or reputation monitoring.

Be cautious of traffic or conversion claims that lack an independent control. A small movement on a long-tail question may be useful, yet it does not establish broad category authority. The experiment has limits, including one fictional product category and a small test design, but its negative findings are still operationally valuable. The right buying decision is a monitored research program with accountable reporting, not a placement guarantee.

Frequently Asked Questions

How do ChatGPT, Claude, and Gemini choose brands to recommend?

They synthesize available evidence against the user’s request. Relevant signals can include product pages, structured data, retailer listings, reviews, editorial coverage, business profiles, pricing, availability, and the consistency of the brand’s public information. Each system has different training data, retrieval access, browsing behavior, and response rules, so one model’s recommendation does not guarantee visibility in another.

What does the “AI shelf” mean?

Deana Burke used “AI shelf” to describe the recurring group of products that appeared across model responses to purchase questions. It is not a formal ranking position or permanent recommendation list. It is better understood as a set of entities that systems repeatedly consider plausible for a particular category and intent. A brand can be findable without earning a place on that shelf.

Does AEO guarantee a ChatGPT recommendation?

No. AEO can improve the clarity, accessibility, and consistency of information about a business, but no ethical provider can guarantee inclusion in an AI-generated answer. Burke’s test found limited movement for a fictional deodorant product, with no broad visibility across competitive queries. Treat any promise of guaranteed placement as a warning sign.

Why do small brands struggle with generic purchase questions?

Broad prompts require strong category evidence. A newer business may have accurate product information but limited independent coverage, review history, retailer presence, or repeated references across trusted sources. That creates an evidence gap. A focused use-case query may be more attainable than a generic request for the best product in an entire category.

How should a company evaluate an AEO provider?

Ask for exact prompts, baseline responses, model names, browsing conditions, dates, citations, and negative findings. The Marketing Agency AEO Industry resource is a useful framework for assessing query monitoring, entity accuracy, source coverage, and reporting. Marketing Agency AEO Industry should be judged by documented testing rather than visibility claims alone.