This piece looks at how quickly ChatGPT’s grip has loosened, why a challenger like Mistral changes the shape of the field, and what a durable approach looks like once you accept that the list of engines you need to appear in will keep changing. The short version is that the winning move is to optimise for humans rather than platforms, then measure whether it is working across every surface that matters.
Ask most marketing teams which AI platform to optimise for and the answer still comes back: ChatGPT. For roughly two and a half years that was sound, because ChatGPT arrived at the end of 2022 and quickly became the whole of AI search, so building a strategy around a single surface was simply the sensible reading of the market. What has changed is that the market stopped standing still. ChatGPT’s share of AI referrals has slipped from around nine in ten towards six, Claude and Gemini have taken real ground, and a well-funded European challenger has started to circle, so the single-surface assumption underneath most AI content strategies has quietly expired. Staying visible while the surfaces keep rearranging themselves comes down to the strength of your content and the breadth of your measurement across them.
Key takeaways
- For over two years, optimising for ChatGPT was effectively optimising for AI search, because it was the one surface that mattered.
- ChatGPT’s share of AI referrals has fallen from around 89% in mid-2025 to around 63% by early 2026, and every serious tracker agrees it no longer holds the whole market.
- Claude has risen to become the clear number two, with Gemini and Perplexity also taking ground.
- Mistral, Europe’s best-funded AI lab, shows the list of engines is still moving, and it carries a data-sovereignty angle that matters for UK and EU buyers.
- Optimising for one platform is fragile, because AI systems build answers from signals across the whole web rather than ranking a single page.
- The durable approach is to optimise for humans, not platforms, through useful content, a clear strategy, a topical content map and clean schema, and then to benchmark visibility across surfaces.
When ChatGPT was the only surface that mattered
When ChatGPT launched at the end of 2022, it did not just lead the AI assistant market, it was the market. For the next two and a half years, most people who used AI to find or summarise information were using ChatGPT, and the phrases “optimise for AI” and “optimise for ChatGPT” pointed at the same job. Goodie’s tracking still put ChatGPT at around 89% of measurable AI referrals across its brand panel as recently as mid-2025, so a ChatGPT-first strategy was not lazy thinking, it was an accurate reading of where the attention sat.
That dominance shaped how agencies and in-house teams built their playbooks, so content was structured to be picked up by one system, success was judged by whether that one system cited you, and the whole approach rested on the assumption that ChatGPT would stay the front door to AI search. That assumption was reasonable at the time, though it has since stopped being true.
The room got crowded
By early 2026, the picture had changed enough to break the old playbook. In its 2026 AI search traffic report, Goodie found ChatGPT’s share of measurable AI referrals had fallen from around 89% in mid-2025 to around 63% by the start of 2026, with Claude rising to become the clear number two and Gemini and Perplexity taking ground behind it. Global trackers tell the same story from a different angle, with StatCounter recording ChatGPT slipping below 80% of AI chatbot referrals for the first time in early 2026, down from a peak above 84% the year before.
The exact split depends on who is counting, the audience they measure, and whether they track referrals or raw usage, so the numbers move accordingly. What every credible source agrees on is the direction: ChatGPT no longer holds the whole room, and the share it has shed has gone to engines that retrieve, cite and recommend in their own ways. A brand that is visible in one of them is not automatically visible in the others.
Europe’s challenger: Mistral
The clearest sign that this reshuffle is not finished is Mistral. The Paris-based lab is Europe’s best-funded AI company, valued at around €11.7bn in a 2025 funding round led by the Dutch chip-equipment maker ASML, and it has built its pitch on open models and European data sovereignty. In early 2025, France’s president publicly urged people to use its Le Chat assistant instead of ChatGPT, which puts political weight behind it as well as commercial.
Mistral barely registers in AI referral data today, and its near-absence is precisely what makes it worth watching, because it shows how quickly a new surface can arrive from outside the current top four. For UK and EU businesses, particularly in regulated sectors where data residency and compliance carry real weight, a credible European assistant can move from marginal to material fast. The engines you need to appear in next year may not be the ones on your dashboard now, which makes any strategy pinned to a fixed list of platforms fragile.
Why optimising for ChatGPT alone is now a mistake
The case against a single-platform strategy goes beyond ChatGPT losing share, because the surfaces absorbing that share are not interchangeable, and neither are the people using them. ChatGPT, Claude, Gemini and Perplexity each run a different retrieval pipeline, apply different citation logic and serve different intents, so the content and signals that earn a mention in one do not automatically earn it in another. A brand tuned to the habits of a single system is quietly opting out of the rest.
AI systems also do not rank pages the way a search engine does; they assemble an answer by drawing on many sources at once and pulling the passages that best support the explanation, which is the deeper reason the old playbook misfires. Chasing one platform’s quirks with keyword-led tactics misreads how citation actually happens, because inclusion is driven by clear, credible information that appears consistently across the wider web rather than by how well a page is tuned to a single interface.
Measurement makes the risk worse, because the referral logs most teams rely on undercount AI badly. Native apps strip the referrer when someone follows a link, so a share of what lands in your “direct” traffic is really AI. Google folds its own AI Overviews into ordinary organic reporting, where they cannot be separated out, so a team optimising only for the platform it can measure is deciding on a partial picture.
If the platform you optimise for lost half its share in the next six months, would your visibility fall with it? For a strategy built on a single platform the honest answer is yes, which tells you the ground it stands on has already proven it can move.
Optimise for humans, not platforms
If the list of engines keeps changing, the smart response is to build the signals that every engine draws on, because those signals stay stable even when the platforms do not. In practice that means optimising for the human asking the question, not the machine parsing the page, and it rests on four things.
Good content that reads like a person wrote it
The starting point is content genuinely useful to the person asking, written in a voice a machine would struggle to reproduce. Distinctiveness has become a ranking signal in its own right, because AI answer engines have little reason to cite a page that reads like every other AI-generated piece on the topic. Content with a clear point of view, original evidence and a human editor behind it earns citations that generic output cannot, and that quality bar is the foundation of any credible AI SEO programme.
A strategy, not a scatter of posts
Useful content still needs direction, so the second move is deciding what your brand should be known for, then covering it with intent. A handful of authoritative, comprehensive pieces on the subjects that matter to your buyers will do more for AI visibility than a stream of thin, keyword-targeted posts, because AI systems judge your content in the context of everything else you have published on a topic. The goal is to be recognised as a genuine source on a defined set of subjects, which is a positioning decision before it is a content one.
A content map that builds topical authority
Depth on a subject is what signals expertise, so the third move is to map your topics deliberately and cover them in a connected cluster instead of isolated one-offs. When your pages reinforce each other and align with how a subject is discussed across the wider web, engines can associate your brand with that subject with more confidence. It is worth being disciplined about scope, and our guide on how to build a topical authority map covers what to include and, just as importantly, what to leave out.
Schema that makes your content machine-readable
The fourth move is technical: clean, accurate schema and a clear content structure so AI systems can parse your pages and extract the exact passage that answers a query. Structured data, sensible headings and self-contained explanations make it easier for an engine to lift and cite you, whichever engine it is. This is unglamorous work, and it is what allows an AI to take a clean, self-contained passage from your page and reuse it with confidence.
These four moves optimise for humans, and because the engines are ultimately trying to serve those same humans, they keep their value across whichever surface rises next. It is also one of the clearest reasons AI still needs people in the loop, for the judgement about what to say, how to say it distinctively, and which subjects to own.
Know where you stand as the surfaces move
Building for humans answers the question of what to do, but it leaves open how you know it is working across a set of surfaces that keeps changing and mostly does not report clean referral data. You cannot manage what you cannot see, and a single referral log both undercounts AI and hides which engines are actually citing you.
This is where measurement widens from one dashboard to a benchmark across surfaces. Tracking your presence in ChatGPT, Claude, Gemini, Perplexity and Google’s AI answers, then watching how it shifts month to month, turns AI visibility from guesswork into something you can manage and report on. It is the reason we built Search Radar, and the reason benchmarking now sits at the centre of any serious AI search programme.
ChatGPT is far from finished, but its years as the only surface worth optimising for are behind us. The brands that stay visible through the next reshuffle will be the ones building for humans and measuring across every engine that matters. If you want to see where your brand stands across the AI surfaces today, and track it as the field keeps moving, our benchmarking reports are built for exactly that.
Frequently asked questions
What is the difference between GEO and SEO?
Generative engine optimisation (GEO) structures content so AI answer engines can cite it, while search engine optimisation (SEO) works to rank pages in traditional search results. The two overlap heavily, because both reward clear, authoritative, well-structured content. SEO aims for a position in a list of links, while GEO aims for inclusion in an AI-generated answer across engines like ChatGPT, Gemini and Perplexity. Most brands now need both.
What is generative engine optimisation, or AI SEO?
Generative engine optimisation, often called AI SEO, is the practice of structuring and writing content so AI-powered search platforms can read, extract and cite it in their answers. It targets the platforms people now use to find information, including ChatGPT, Perplexity, Gemini and Google’s AI Overviews. In practice it combines clear writing, topical depth, entity clarity and clean schema, so that when an engine assembles an answer, your content is one of the sources it can confidently reuse.
Should I still optimise my website for ChatGPT?
You should optimise for ChatGPT and the other major AI engines, not for ChatGPT on its own. It remains the largest single source of AI referrals, so it still deserves attention, but its share has fallen from around 89% in mid-2025 to around 63% by early 2026 while Claude, Gemini and Perplexity have gained ground. Optimising for ChatGPT alone leaves you invisible to everyone using the others, so the stronger approach builds content and authority that work across all of them.
How do I get cited by AI?
You get cited by AI by publishing clear, credible content that consistently explains the topics you want to be known for, and by making it easy for engines to extract. AI systems assemble answers from sources that appear reliable across the wider web, so citations follow entity recognition, topical depth, original evidence and clean structure. Being referenced by other credible sites and publications reinforces this. Content that reads like generic AI output rarely gets picked, because engines have no reason to prefer it.
How do I appear in Google AI Overviews?
You appear in Google AI Overviews by being a clear, authoritative source that Google’s AI can extract to answer a query. Overviews are powered by Gemini and now appear on a large share of searches, and the pages they cite are not always the top organic result. Structured content, direct answers to specific questions, topical authority and strong trust signals all improve your chances. Traditional SEO foundations still matter, because AI Overviews draw on Google’s existing index.
How do I track my visibility across AI platforms?
You track AI visibility by testing and monitoring your brand’s presence across each major engine, because referral analytics alone undercount it. Native apps strip referrers and Google bundles AI Overviews into organic traffic, so a single dashboard misses most of the picture. A cross-surface benchmark that checks how often you are cited in ChatGPT, Claude, Gemini, Perplexity and Google’s AI answers, and how that changes over time, gives a far more accurate read. This is the job benchmarking tools are built for.
