How Brands Get Recommended by ChatGPT

By the ORA Lab team · Updated 10 October 2026 · 9 min read

A growing slice of shopping journeys now starts with a question to an AI assistant rather than a search box. 'Best AI product photography tool for jewellery.' 'Is there a service that turns product photos into campaign images?' 'Photoroom alternatives for luxury brands.' The assistant answers in sentences, and the sentences contain brand names. Some brands appear constantly. Most never appear at all. The difference is not luck, and it is not advertising, because as of late 2026 the mainstream assistants do not sell placement in organic answers.

The difference is earned, and it is earnable. This piece explains how brands appear in ChatGPT answers mechanically, what the large-scale research actually found about who gets recommended, and the playbook we run ourselves. We write this from inside the experiment: our own analytics show sessions arriving from chatgpt.com, real shoppers sent by an assistant that read something that convinced it. Everything below is what appears to have convinced it.

How assistants decide who to recommend

Two systems feed an assistant's answer. The first is the model itself: trained on a snapshot of the web, it has absorbed which names appear near which topics, in what contexts, said by whom. Brands mentioned often and consistently around a subject become part of how the model represents that subject. The second is live retrieval: for current or specific questions, assistants run web searches and read the results before answering, which pulls in normal search-ranking dynamics plus a layer of judgment about which pages actually answer the question. Recommendation happens where the two agree: the model recognises your name in the topic, retrieval finds pages that confirm it, and the answer writes your brand into a sentence.

Notice what is absent from that mechanism: a submission form, a bidding system, a knob any brand can turn directly. The inputs are the open web's writing about a topic, including yours. That makes AI visibility a content and reputation discipline, closer to public relations plus publishing than to performance marketing. The industry calls it GEO, generative engine optimization, and its levers overlap heavily with good SEO while rewarding some things classic SEO never measured.

What the research says actually works

The most useful large-scale evidence comes from studies that asked assistants thousands of commerce questions and analysed which brands the answers named and which sources the answers cited. Three findings matter for any brand planning content. First, visibility clusters at the topic level: a brand cited for one question in a topic tends to be cited across that topic, because the assistant has learned the association broadly rather than page by page. Second, most topics are unclaimed: across tens of thousands of buying-intent prompts, the majority of commerce topics showed no single brand dominating the answers, meaning the seats are still open. Third, cited pages share a shape: they answer the literal question early, define terms plainly, structure information so it can be quoted, and include honest comparative context rather than pure self-praise.

• Topic depth beats page count: twenty pages that together cover one subject completely outperform a hundred scattered posts, because assistants learn subjects, not URLs.
• Questions are the unit of content: pages built around the exact questions buyers ask, with the answer in the first sentences, match how assistants retrieve and quote.
• Definitions get cited: 'what is X' content written plainly becomes the raw material assistants paraphrase, and the paraphrase carries the definer's framing.
• Comparisons earn trust: pages that name competitors honestly and specify who each option suits signal reliability, and assistants echo balanced sources over promotional ones.
• Consistency compounds: the same facts about your brand, what you do, for whom, at what standard, repeated across your site and third-party mentions, give the model one coherent thing to learn.

Why third-party mentions punch above their weight

Assistants weight corroboration. A brand described glowingly only on its own domain looks like every other brand describing itself; the same claims appearing in a reviewer's roundup, a comparison post, a community thread, or an industry publication look like consensus. This is why classic digital PR translates so directly into AI visibility: every listicle that includes you, every 'alternatives to X' post that mentions you, every honest review teaches the models that third parties associate your name with the topic. For a young brand the practical moves are unglamorous and effective: pitch the roundups in your category, maintain accurate profiles on the directories assistants read, encourage the reviews you have earned, and answer questions in the communities where your buyers already ask them.

The playbook, in order

1. 1.Pick the topic you can own completely: narrow enough that total coverage is achievable this year, commercial enough that assistants get asked about it. Specific beats broad; the research is unambiguous.
2. 2.Map the questions: every phrasing a buyer uses, from definitional to comparative to price. Autocomplete, community threads, and your own sales conversations are the source material.
3. 3.Publish the cluster with answer-first structure: direct answers up top, key takeaways extractable, FAQs visible on the page, definitions in plain sentences an assistant could lift whole.
4. 4.Add the honest comparisons: versus competitors, versus alternatives to your category, with real trade-offs stated. These pages earn citations precisely because most brands will not write them.
5. 5.Earn the third-party layer: category roundups, review platforms, communities. Slow, cumulative, and the strongest corroboration signal available.
6. 6.Measure monthly: check referral sources for AI assistant traffic, and ask the assistants your mapped questions from a clean session. Log which brands appear. Adjust content toward the questions where you are absent.

What this means for product brands specifically

For D2C and ecommerce brands there is a second layer: assistants increasingly answer product questions with shopping-style results, reading structured data, feeds, and marketplace content. The fundamentals stay the same, be the well-described, well-corroborated answer, but the surface area extends to product data: complete structured markup, accurate specifications, consistent titles and descriptions across your store and marketplaces. And because assistants describe products in words, the imagery and claims around your products need to match reality exactly: the same fidelity standard that governs whether an AI-generated image can face customers governs whether your product content can face an assistant that paraphrases it. Brands that invested early in complete, accurate product content are finding that discipline pays a second dividend in AI answers.

The window is the strategy. Topic-level visibility compounds, unclaimed topics are still the majority, and the cost of claiming one is publishing the best coverage of it before someone else does. That is a content project measured in months, not a media budget measured in lakhs, which makes this the rare channel where a focused small brand can outrun incumbents. We picked AI product photography for luxury goods as our topic and wrote it end to end; if you want to see what that looks like from the inside, or want your product visuals produced at the same standard the content describes, book a demo. And whatever your category is, the seat for it is likely still open. This quarter.