ChatGPT and Gemini Agree on Only 63 to 70 of the Top 100 Brands

ChatGPT and Gemini do not surface the same brands, even when they are answering similar commercial questions. Semrush found that major platforms shared only 63 to 70 of their top 100 mentioned brands in the UK, which means a single "AI visibility" score can hide more than it reveals.¹

The more marketing moves through assistants, agents and automated platforms, the less useful it becomes to think of visibility as one ranking. A brand can be well represented in Google, barely present in ChatGPT, described differently in Gemini, and still have automated media systems spending money on its behalf. The week's more interesting stories all point to the same pressure: brands need to make their source material, decisions and identity clearer because more software is now interpreting them.

There is no single answer

Semrush's UK AI Visibility Index makes the fragmentation hard to ignore. Any two major platforms shared only 63 to 70 of their top 100 most mentioned brands in the UK, compared with 70 to 78 in the US.¹ That is not a small statistical wrinkle. It means marketers cannot assume that being prominent in one assistant will carry neatly into another.

The source mix changes too. In Semrush's UK sample, news sites made up an average 29.5% of citations among the 100 most-cited domains on ChatGPT, compared with 19.6% in the US.¹ A brand's visibility can therefore depend on the model, the market, the question and the third-party sources available to support the answer. That makes the source mix part of the strategy rather than a reporting footnote. One score cannot capture all four variables without flattening the useful detail.

That matters because the commercial questions are getting sharper. Fintel Connect's analysis of 2,400 answers found that around 90% of decision-stage prompts produced a citation naming a brand, compared with 8.8% at the awareness stage, while more than 54% of recommendation answers cited an affiliate page.² The customer asking "what is a current account?" is in a different moment from the customer asking "which current account is best for me?", and the assistants treat those prompts differently. A useful AI in marketing strategy has to start with the buyer question, not with a vague goal to "rank in AI".

A ranking can still be useful, but only when the unit of analysis is clear. A marketer looking at UK financial services, a US ecommerce brand and a B2B software company are not solving the same discovery problem, even if a vendor gives each of them a visibility percentage. The useful comparison is prompt by prompt, market by market and source by source, because that is where the reasons for a recommendation become visible. Otherwise the metric risks rewarding activity without explaining whether the business is present at the moment a buyer is actually choosing.

This also changes how teams should diagnose a visibility problem. ROI Amplified's new audit checks whether ChatGPT, Gemini and Claude mention, cite or recommend a brand for specific buyer questions, then compares the result with named competitors.³ That approach is more useful than chasing an overall score because it exposes the exact prompts where the business disappears. The operational question is not "how visible are we?" but "which buyer questions are we absent from, and what evidence is the assistant using instead?"

The website is becoming source material

For years, digital marketing treated the website as the destination. AI assistants are turning parts of it into source material that may be read, summarised or acted on before a person visits. Dfinitiv's SmartCXP AI is an early example: loyalty programmes can expose branded experiences inside ChatGPT and Claude through MCP, letting customers discover offers and benefits while staying inside the assistant.⁴ The brand still owns the offer and rules, but it may no longer own the screen where the customer encounters them.

Contentsquare is measuring the other side of that change. Its new Experience Platform separates LLM and autonomous-agent activity from human sessions, tracks AI-referred traffic and includes agent interactions in attribution.⁵ That means the visitor a marketing team needs to understand may be software acting for the customer rather than the customer directly. The old funnel assumed the brand owned most of the important surfaces between discovery and conversion. That assumption is weakening.

The immediate temptation is to treat this as another distribution channel and produce more content for it. That would be a poor response. If assistants are assembling answers from a mix of first-party pages, news, affiliate sites and other sources, flooding the web with interchangeable text gives them very little reason to distinguish one business from another. The better job is to make the business easier to describe accurately: clear product information, specific claims, consistent language, useful structured data and credible third-party references.

This puts more pressure on the content supply chain behind the answer. A product page written for search, a press story written for a journalist and a social post written for followers may all describe the same offer with different emphasis. That is normal, but the underlying facts should agree. If price, eligibility, product names or positioning drift across those sources, assistants are being asked to reconcile contradictions the brand created itself.

This is where Instagram content strategy and AI search start to overlap in a way marketers may not expect. A recognisable brand cannot live only in a design system or a social feed if assistants are also forming an opinion from the wider web. The language used in product pages, press coverage, customer education and social posts needs enough consistency that different systems arrive at a similar understanding. Brand identity is becoming something software has to be able to read, not merely something people can recognise visually.

Easier setup makes the brief more valuable

Google is testing campaign creation that begins with a sentence instead of a settings screen. Advertisers describe the result they want, and Google suggests goals and assets; separate tests are also giving the system more freedom over bidding and audience expansion.⁶ This lowers the amount of platform knowledge required to get a campaign running, which is useful for smaller businesses without a dedicated PPC specialist. It also means more of the campaign is being inferred from fewer words.

It also makes weak instructions more expensive. When campaign mechanics are hidden behind natural language, the person writing the brief has to be clearer about the audience, the commercial outcome and the constraints that should not move. A vague sentence can now travel further because the platform can turn it into more decisions. Ease of use shifts value away from knowing where every setting lives and towards knowing what the business is trying to achieve.

The same pattern appears in less glamorous work. Tesco Media says conversational audience recommendations can cut segment building from five to ten minutes per segment to under a minute for multiple segments.⁷ Guideline's ingestion agent reads differently formatted Excel and CSV media plans, maps their fields, flags uncertain matches and waits for a planner to confirm them before saving.⁸ Those tools are not mainly replacing judgement. They are removing the manual translation work around it.

That distinction is useful because "human in the loop" often gets interpreted as a person manually touching every stage. It should mean a person is present at the points where judgement changes the outcome. In Guideline's case, the planner does not need to re-enter every row, but they do need to confirm the uncertain mapping before anything is committed.⁸ That is a much clearer responsibility than asking a person to watch every automated step. Good automation removes clerical effort while making the decision boundary easier to see.

The same principle should apply to Instagram AI content management and AI captions for Instagram business posts. A system can assemble drafts, suggest formats, find source assets and adapt copy, but the business still needs to decide what claim it is comfortable making and whether the post sounds like the brand. Faster content production is useful only when the brief and source material are specific enough to survive the automation.

Cheap output raises the value of proof

The cost of creating and testing material is falling quickly. ZoomRx launched a marketing asset testing product for pharma teams that combines unlimited synthetic testing with roughly six research cycles using real healthcare professionals.⁹ That structure is revealing because the synthetic layer is used for speed, while real people remain part of the validation. The company is not treating simulated feedback as the final authority.

Getty Images is making a related bet from the asset side. Its planned take-private transaction is backed by up to $533 million of new equity, and Getty is emphasising its premium archive plus generative tools trained on permissioned content with commercial usage rights.¹⁰ When image generation is cheap, knowing where an asset came from and whether it can be used safely becomes a product feature. Provenance stops being back-office paperwork and becomes part of the value proposition.

Audience expectations are moving the same way. SiriusXM Media's 2026 study says 76% of podcast listeners are active generative AI users, yet 87% want disclosure when an advert contains AI-generated content or voices, and 91% want companies to disclose when they use AI.¹¹ People can be comfortable using these tools themselves while still wanting brands to be explicit about synthetic media. Adoption does not cancel the expectation of honesty.

Google has also tightened the operational side by making its guidance for generated content more explicit. Publishers are expected to manually fact-check and review generated titles, meta descriptions, structured data and image alt text before publishing.¹² That is a sensible standard because these small fields increasingly feed both search engines and assistants. A wrong title or fabricated description does not stay tucked away in the CMS. It can become the raw material for another system's answer.

Distinctiveness has to survive translation

Marketing teams used to think mostly about what a person saw. Now they also have to think about what an assistant inferred, what an agent was allowed to do, what an automated media platform optimised and which source another model trusted. Those systems will not interpret a brand in exactly the same way, so the aim cannot be perfect consistency across every platform. The aim is to give each system enough accurate, distinctive material that the business still comes through recognisably.

That makes generic AI content particularly expensive. If hundreds of businesses publish the same competent language, use the same stock claims and allow the same optimisation systems to decide what gets shown, they remove the evidence that makes one brand easier to remember and easier to recommend. AI content marketing works best when it starts with real source material, specific products, real customer needs and language the business would be willing to stand behind. The machine can help with scale, but it cannot invent distinctiveness without making something up.

There is a practical test for teams using Instagram marketing AI, paid-media automation or assistant-facing content. Take the five commercial questions a customer asks close to a decision, then check how different assistants describe the business and which sources they cite. Next, inspect the internal workflow and mark where software can act without approval, where it must show its work and where a person remains accountable. That exercise connects visibility, content quality and automation in a way a dashboard score cannot.

The week's stories are a reminder that more capable software does not reduce the need for judgement. It moves judgement earlier, into the brief, the source material, the permission boundary and the final review. Brands that treat those as operational details will automate faster and become less recognisable. Brands that treat them as part of the marketing system will give both people and machines something clearer to work with.

Sources