How ChatGPT Decides Who to Recommend (And What You Can Do About It)

The question wording decides the category

One pattern worth knowing: ChatGPT's decision about which brands to mention is framed by the category it slots the question into. A brand can be near-absent under generic category wording and prominent under the specific wording its buyers use, with nothing else changing. ChatGPT's mental model of which brands belong in each category is sharply boundaried. If you measure on the wrong category, the right brand looks invisible. If you build content for the wrong category, you optimise for a slot the AI does not put your brand in.

The Shift Already Happened. Most Brands Missed It.

In a consumer survey reported by Search Engine Land in January 2026, 37% of consumers said they now start searches with AI tools instead of Google. It is self-reported, but it points one way.

When someone asks ChatGPT "what's the best project management tool for remote teams" or "which travel company should I book a luxury safari with," the response is not a list of ten blue links. It is a recommendation, naming one, two, maybe three brands, described and positioned against each other.

Your brand is either in that answer or it is not.

The rules of visibility have changed. And most marketing teams are still optimising for a game that is rapidly shrinking.

How ChatGPT Actually Decides

ChatGPT does not rank pages. It does not crawl your website the way Google does. It draws from two separate systems, and understanding the difference matters.

Training data is everything the model absorbed before its knowledge cutoff. Every web page, every Reddit thread, every review, every press article, every academic paper it was trained on. This forms the model's baseline understanding of your brand: what you do, who you serve, how you compare to competitors. If your brand was rarely mentioned in the training corpus, ChatGPT may not know you exist.

Real-time search is the newer layer. When ChatGPT browses the web (via Bing), it supplements its training data with live results, so recent content, fresh reviews and current press coverage can make a difference here. ChatGPT does not simply surface the top Bing result. It synthesises. It reads multiple sources, cross-references them, and constructs an answer that reflects the consensus view.

This means ChatGPT is not deciding based on one signal. It is building what I think of as an entity model: a composite picture of your brand assembled from thousands of data points across the internet. How consistently you are described. How often you are mentioned by others. How authoritative those others are.

What your brand says about itself carries far less weight than what others say about your brand. In our testing, third-party sources, including Reddit, review sites, industry publications, and comparison articles, dominate ChatGPT's citation patterns. Your own website is one input among hundreds.

The Algorithm Change Nobody Noticed

On 1 December 2025, something significant changed inside ChatGPT's search behaviour. Seer Interactive published analysis showing that citations per response jumped 81%, from an average of 5.7 sources to 10.4 sources per answer. From around that date, across the 43 brands Seer tracks, answers referenced nearly twice as many sources on average.

More citations means more opportunities to appear - but also more competition for each answer. The bar for getting cited did not necessarily drop. The number of brands that could appear in a single response simply increased.

For brands that had strong third-party presence, this was a net positive. More citation slots, more chances to be included. For brands relying solely on their own website and a few backlinks, the change made them relatively less visible. The signal-to-noise ratio shifted.

We saw this in our own data. Brands with broad third-party coverage, mentioned on Reddit, reviewed on comparison sites, featured in industry press, saw their ChatGPT appearance rates increase in Q1 2026. Brands with thin third-party footprints saw theirs stay flat or decline, even when their Google rankings held steady.

What Our Data Shows

Our study of 100 UK buying questions across six AI platforms (June 2026) ran every question through ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and Google AI Mode in grounded mode. We run each question several times on the chatbot platforms because AI responses are non-deterministic: ask the same question twice and you may get different brands recommended.

Here are the patterns that stood out.

FAQ and list-based content are among the most frequently cited structures. Research from Previsible found that lists, FAQs, and comparison tables are the structural elements AI models cite most often. When a brand publishes clear, direct answers to the questions their customers ask, and those answers are structured in a way that AI can parse, they appear in responses far more often. In our testing, queries framed as questions ("what is the best...", "how do I choose...", "which company should I use for...") pulled structured, answer-format content at significantly higher rates than long-form prose.

Third-party sources dominate over brand websites. Across the major AI platforms, the vast majority of cited sources are not brand-owned. Reddit threads, Trustpilot and G2 reviews, industry comparison articles, and publisher roundups account for the bulk of what AI models reference. Your own website may appear as a secondary source, but the primary recommendation usually comes from what others say about you.

Each platform behaves differently. This was one of the more striking findings. The same brand can be named often on one engine and rarely on another for the same questions. Perplexity, which stopped testing advertising in February 2026, often surfaces completely different brands than ChatGPT or Claude. Google AI Overviews draw on Google's own index. Claude runs its own web searches when it needs current information, and in our six-platform study it named the most brands per answer. Treating "AI search" as one monolithic channel is a mistake.

Brand search volume correlates with recommendation rate. Brands that people actively search for by name, indicating broader awareness, tend to be recommended more often by AI models. This makes sense: if millions of people search for a brand, there is a large volume of content about that brand across the internet, which feeds the training data and the real-time search layer.

Consistency of brand description matters. When a brand is described the same way across multiple sources, with the same positioning, category language, and value proposition, AI models appear to have higher confidence in recommending it. Brands with inconsistent messaging across their website, reviews, directory listings, and press coverage are harder for AI to categorise, and harder to categorise means harder to recommend.

Why Google Rankings Do Not Translate

The most important pattern in this data: a brand can rank number one on Google for a target keyword and score 0% across every AI platform for the same query.

I have seen this repeatedly, not as an edge case but as a pattern.

The reason is structural. Google ranks pages. It evaluates individual URLs based on relevance, authority, and technical signals, then orders them in a list. AI models do not rank pages. They synthesise answers from a broad corpus of information and recommend brands (not URLs) based on entity-level signals.

A page can be perfectly optimised for Google, with a strong title tag, fast load time, good internal linking, and solid backlink profile, and still contribute nothing to AI visibility if it is the only place on the internet that makes a case for the brand.

A study by Loamly across 2,089 brands found that 77% are effectively invisible to AI platforms, and that the biggest predictor of AI visibility is off-site authority, not website optimisation. In our own data, we regularly see brands with strong organic positions that are completely absent from AI responses, and brands with modest Google rankings that appear consistently across ChatGPT, Claude, and Perplexity.

To be clear: strong SEO still matters. Structured data, topical authority, and quality content all contribute to how AI models understand your brand. SEO is the foundation. But a foundation alone is not a building. If your AI search strategy is "keep doing SEO and nothing else," you are optimising for one platform while your customers are spread across six.

The 5 Signals That Matter

From our research and the published studies cited here, these are the five signals that most often go with a brand appearing in AI recommendations.

1. Third-Party Mentions

Third-party mentions are the biggest lever: the volume, quality and recency of what others say about your brand across the internet.

Reddit threads where your brand is mentioned or recommended carry significant weight. So do review sites (Trustpilot, G2, Capterra), industry comparison articles ("best X for Y"), press coverage, and expert roundups.

What to do: Audit your third-party footprint. Search your brand name on Reddit, review sites, and industry publications. If you find gaps, build a citation strategy: encourage reviews, engage authentically on Reddit, pitch data-driven stories to industry press, and get listed on relevant comparison directories.

2. FAQ and Question-Answer Content

AI models are built to answer questions. Content that is structured as direct answers to specific questions is inherently easier for them to parse and cite.

This is not about stuffing FAQ schema onto your homepage. It is about publishing genuine, useful answers to the questions your customers actually ask, on your website, on third-party platforms, and in formats that AI can extract.

What to do: List the 20 questions your customers ask most often. Publish clear, direct answers on a dedicated FAQ page and within relevant blog content. Use FAQ schema markup. But also answer those questions on Quora, Reddit, and industry forums. The third-party layer matters more than the on-site layer.

3. Entity Consistency

When every source on the internet describes your brand the same way, with the same category, positioning, and key differentiators, AI models can recommend you with confidence. When your LinkedIn says one thing, your website says another, and your directory listings say a third, the model has low confidence and may skip you entirely.

What to do: Audit how your brand is described across your website, Google Business Profile, LinkedIn, directory listings, review sites, and press coverage. Align the language. Use the same category terms, the same positioning statements, the same key phrases. This is about consistency, not keyword stuffing.

4. Brand Search Volume

Brand search volume is the one signal you cannot fake quickly. When people search for your brand by name, it creates a feedback loop: more branded searches means more content about your brand, which means AI models have more data to draw from, which means higher confidence in recommending you.

What to do: Brand building matters. PR, partnerships, events, thought leadership, and advertising all contribute to branded search volume. This is the long game, but there are also tactical moves: co-marketing with established brands, contributing to high-visibility publications, and creating data or research that gets cited widely.

5. Content Freshness and Recency

ChatGPT's real-time search layer favours recent content. Google AI Overviews heavily weights freshness. A blog post from 2022 carries less weight than one published this month, particularly for queries where the landscape is changing.

What to do: Update your key content regularly. Refresh statistics, add new examples, and republish with current dates. Publish timely commentary on industry changes. AI models are increasingly incorporating recency as a ranking factor in their real-time search components.

What You Can Do This Week

Forget the 90-day roadmap for a moment. Here are five things you can do in the next five days that will start shifting your AI visibility.

Monday: Run the test. Open ChatGPT, Claude, and Perplexity. Ask the 10 questions your customers ask most often, the ones that should lead to your brand. Record which brands appear. Record which ones do not. You now have a baseline.

Tuesday: Audit your third-party presence. Search your brand name on Reddit. Check your Trustpilot, G2, or industry-specific review profiles. Look at comparison articles in your category. Count how many third-party sources mention your brand. Compare that number to your top competitor.

Wednesday: Publish five FAQ answers. Take the five most common customer questions and publish clear, direct answers on your website. Add FAQ schema markup. Keep each answer under 150 words. Make them factual, not promotional.

Thursday: Claim your directory listings. Identify the five most relevant directories or comparison sites in your industry. Ensure your brand is listed, described consistently, and up to date. If you are not listed, submit your profile.

Friday: Ask for three reviews. Reach out to three satisfied customers and ask them to leave reviews on the platforms that matter most in your category: Google, Trustpilot, G2, or industry-specific sites. Third-party reviews are one of the highest-signal inputs for AI models.

None of this requires a consultant or new technology. It requires focus on the signals that AI actually uses, instead of the signals that Google used to reward.

The Bigger Picture

The advertising layer has arrived. ChatGPT Ads launched in the US in February 2026 and have since reached the UK and 31 European markets. OpenAI says ads do not influence ChatGPT's answers, but the space around the answer is now paid.

But organic recommendations are not going away. Even with ads, the core answer is still generated from the model's understanding of the category. Ads appear alongside recommendations, not instead of them. And Perplexity's decision to drop advertising entirely in February 2026 shows that not every platform is moving in the same direction.

The brands that invest in AI visibility now, while the playbook is still being written and most competitors are still focused exclusively on Google, will have a structural advantage that is very difficult to reverse-engineer later.

The data has already settled whether AI search matters. What is still open is whether your brand appears when your customers ask.