Why AI Recommends Your Brand in One Category but Ignores It in Another

I recently read Maryanna Franco's article, How category framing changes which brands AI recommends, which draws on research she conducted with Joao da Silva.

It puts data behind a problem I think many brands will miss when measuring AI visibility.

Your brand can be well known by AI and still disappear when a customer uses a slightly different category word.

This is not necessarily because AI thinks your brand is weak. It may be because AI does not associate your brand with the category being asked about.

That distinction matters because the diagnosis determines what you should fix.

If AI does not believe your brand belongs in the category, adding more testimonials or product benefits may not help. You first need to make the category association clear.

If AI already includes your brand but recommends a competitor instead, you have a different problem. You need stronger evidence that your brand is the better choice.

In short, your brand must first belong in the consideration set before it can compete to win the recommendation.

What the research found

The Search Engine Land article reports a seven-day test involving 14,140 runs across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews.

In one comparison, changing the category in the prompt from "athleisure" to "athletic footwear" changed the recommendation results substantially:

• New Balance moved from 1% visibility to 90%.
• Lululemon moved from 90% visibility to 0%.
• Nike remained visible across both category framings.

The underlying open-access paper, Beyond Knowledge-Graph Strength: How Category Coding Drives Brand Visibility in Generative AI, reports the wider controlled study using athleisure and sportswear prompts.

The paper distinguishes between two outcomes:

• Recognition: AI can identify and describe the brand when it is named.
• Recommendation: AI introduces the brand without being prompted when answering a category-level question.

The study found that Knowledge Graph strength was more useful for explaining visibility within an established category than across adjacent categories. A brand could be strongly recognised but still fail to appear when the category framing changed.

The paper proposes that this is influenced by two things:

1. How the brand is categorised in structured entity descriptions.
2. The third-party articles, reviews, comparisons and roundups that repeatedly associate the brand with a particular category.

The paper is an open empirical deposit rather than peer-reviewed research, so I would treat the findings as a strong diagnostic hypothesis that deserves further replication. But the practical signal is clear enough to test: category wording can materially change which brands appear.

My view: AI recommendation has two separate stages

The research explains why category language matters. I would extend that into a two-stage diagnostic model.

Stage 1: Does AI believe your brand belongs here?

Before AI compares your brand with competitors, it needs to retrieve your brand as a plausible member of the category.

For example, AI may confidently understand that New Balance is associated with running shoes and athletic footwear. That does not automatically mean it will retrieve New Balance when someone asks about athleisure.

The buyer may see those categories as closely related. The AI system may have learned them from different sets of entities, sources and comparisons.

If your brand is missing at this stage, you have a category-association gap.

Stage 2: Does AI have enough reason to recommend your brand?

Appearing in the category is not the same as winning the recommendation.

Once AI includes your brand in the possible choice set, it still needs enough evidence to decide whether your brand fits the buyer's requirements.

That evidence may include:

• Product suitability
• Relevant features
• Customer reviews
• Expert comparisons
• Demonstrated experience
• Pricing or value
• Availability
• Location
• Use-case fit
• Independent corroboration

If your brand appears but is not selected, you are more likely dealing with a recommendation-proof gap.

Your brand has to belong before it can win.

Is this a category problem or a recommendation problem?

Do not move directly from "AI is not recommending us" to "we need more content."

First, identify where the failure occurs.

This prevents you from applying the wrong solution.

A category-association gap will not necessarily be solved by adding more conversion copy.

A recommendation-proof gap will not necessarily be solved by changing your schema or brand description.

The work needs to address the earliest point where the brand stops progressing towards the recommendation.

How should you test category language?

Start with the buyer's job, not a list of keyword variations.

A person asking for "athleisure brands" and a person asking for "athletic footwear brands" may both be looking for something suitable for everyday exercise. But the category language can activate a different set of brands and sources.

A useful test should therefore keep the buyer situation stable while changing the category framing.

1. Define the buyer decision

Be clear about what the customer is trying to choose.

Find a brand suitable for comfortable everyday training and casual wear.

2. List the category variations customers may use

These could include:

• Athleisure
• Activewear
• Sportswear
• Athletic apparel
• Performance clothing
• Athletic footwear

Do not test every possible synonym. Focus on categories that reflect real customer language and commercially relevant positioning.

3. Keep the rest of the prompt consistent

Change the category wording without changing the buyer, geography, budget or use case.

This gives you a cleaner comparison.

4. Test repeatedly across platforms

Run the prompts across the AI systems that matter to your audience.

One answer is an observation, not a reliable pattern. Repeat the test and keep each platform's results separate before looking for cross-platform conclusions.

5. Inspect the full answer

Do not only record whether the brand was mentioned.

Check:

• Was the brand recommended or merely referenced?
• Which competitors appeared?
• What reason was given for the recommendation?
• Which pages or publications were cited?
• Was the brand represented accurately?
• Did the answer change across repeated runs?

You are not trying to create a new form of keyword ranking. You are testing whether AI consistently connects your brand to a commercially important buyer situation.

What should you change when a category gap appears?

First, confirm that the category is one your brand should legitimately compete in.

Not every adjacent category is worth pursuing. The category should reflect your actual offer, your intended positioning and the way your customers make decisions.

If the category matters, review the full evidence system around it.

Clarify the category on owned properties

Your website should make it easy to understand:

• What category the brand belongs to
• Which products or services support that position
• Which audiences and use cases the brand serves
• What evidence supports the claim

This may involve your homepage, About page, category pages, product pages, internal links and structured data.

But changing one field or adding schema is unlikely to be enough on its own.

Strengthen third-party corroboration

AI systems do not learn your position only from what you say about yourself.

Review whether relevant third-party sources place your brand in the target category:

• Industry publications
• Product reviews
• Editorial comparisons
• Category roundups
• Expert recommendations
• Directories and marketplaces
• Customer discussions
• Partner or association websites

The objective is not simply to generate more mentions. You need credible sources to discuss your brand in the category and buyer context you want AI to understand.

Build recommendation evidence separately

Once the brand is appearing in the category, check whether AI has enough evidence to select it.

This may require stronger:

• Comparisons
• Customer outcomes
• Reviews
• Case studies
• Product specifications
• Expert validation
• Suitability explanations
• Objection handling

Category evidence helps AI understand where you belong. Recommendation evidence helps AI explain why you should be chosen.

Why one AI visibility score is not enough

An overall visibility score can make the brand look healthy while hiding a complete category-level blind spot.

Imagine that your brand performs well across general brand prompts and its established product category. The average may look positive.

But if customers are increasingly using a newer or adjacent category term, the same brand could be missing from the exact conversations that represent future demand.

This is why prompt tracking should be organised around stable buyer associations, not treated as a collection of isolated questions.

For every important association, track:

• Buyer need
• Category or use case
• Platform
• Prompt version
• Brand presence
• Recommendation presence
• Competitors
• Supporting sources
• Answer accuracy

You should also keep controlled benchmark prompts separate from real customer-language prompts.

The controlled panel tells you whether performance is changing under stable conditions. Real customer language tells you whether the panel still reflects how the market actually asks for help.

You need both.

The practical takeaway

The important question is no longer only:

Is AI recommending our brand?

You also need to ask:

Under which category does AI recommend us, and is that the same category our customers use?

If the brand is absent, determine whether AI fails to recognise the entity, connect it to the category or justify the recommendation.

Then fix that specific gap.

Do not assume every visibility problem needs another article. The solution may be clearer category positioning, better product information, stronger independent coverage, more buyer-fit proof or a more reliable measurement panel.