In March 2023 ChatGPT held 88.5% of the AI assistant market. The industry declared it the end of Google, the end of search as we knew it, the beginning of a new era in which everything needed to be rebuilt around the new dominant platform. Agencies rebranded. Consultants pivoted. The phrase "optimise for ChatGPT" entered the vocabulary.
By May 2026 ChatGPT held 46.4%. It lost 42 percentage points in three years. The glass that was described as so dramatically full turned out to contain rather fewer fish than the coverage suggested.
The Gemini figure looks dramatic. It is. But before drawing strategy from it the question worth asking is: what exactly is Gemini?
What "Gemini" actually means
Gemini is a brand name covering several different things and the figures quoted in coverage tend to blend them without distinguishing between them.
The Gemini app — the standalone chatbot at gemini.google.com and the mobile app. This is the direct ChatGPT competitor. 900 million monthly active users at Google I/O 2026. This is the most legitimate like-for-like comparison figure.
AI Overviews — Gemini-powered answers inside Google Search. 2.5 billion monthly users. These are people who searched on Google and received an AI-generated summary. Most of them did not consciously use Gemini. They just searched for something.
Gemini in Workspace — Gmail, Docs, Sheets, Slides. When it drafts your email reply or summarises a document. Counted in some MAU figures and silently absent from others.
Gemini on Android — baked into the device as the assistant replacement. People who have never chosen Gemini are using it every time they press the assistant button.
The Apple deal — January 2026, Apple confirmed Gemini powers the next generation of Siri across 2 billion Apple devices. Not yet fully reflected in most published figures. Coming.
ChatGPT users mostly chose to be there. Most Gemini "users" encountered it because Google placed it in their path. That is not a criticism — it is a description of fundamentally different growth strategies. But it matters when reading the numbers. Gemini's growth is almost entirely a distribution story, not a quality story. Google reduced its Gemini serving costs by 78% in 2025 and put the product inside Search, Chrome, Workspace, YouTube, Android and now Siri. The market share followed inevitably.
The 93% zero-click question
AI Mode — Google's full AI search experience that replaces the traditional results page with a conversation — produces a 93% zero-click rate for traditional organic search. Users who interact with AI Mode do not visit external websites in 93% of cases.
This figure lands badly in most coverage and the reaction is usually either panic or denial. It deserves a more precise reading.
The 93% is almost certainly skewed toward informational queries. "What is CRPS?" "How does PageRank work?" "Who invented the telephone?" These were never high-converting visits. The people asking them were not in a buying mindset. Getting fewer of them through the door does not damage most businesses in any measurable way. These were the fish that were never really in the glass.
The more interesting question is what happens to the 7% who still click through. If AI Mode filters out all the casual informational browsers and delivers only the users who had a specific enough need to go beyond the AI answer, that 7% could be considerably more valuable than the old 100%. Higher intent. Further down the funnel. Already past the research phase. A smaller number of better visitors is not necessarily worse — it depends entirely on what you were doing with the informational traffic in the first place.
The measurement problem is that you cannot currently verify this either way. Google has not provided the attribution layer that would let you see whether the 7% who clicked converted at a higher rate than your previous organic visitors. You can infer from engagement metrics but not isolate the cause. The data belongs to the platform doing the measuring — which is, again, Google.
I want to optimise for GEO
This is the question arriving in briefs and LinkedIn posts in increasing volume. Before answering it, three prior questions deserve a hearing.
Why — what are you actually trying to achieve? If the goal is brand visibility inside AI-generated answers, that is a legitimate objective. If the goal is converting visits into revenue, you may be optimising for something that does not directly produce that outcome. Being cited in an AI Overview is not the same as receiving a visit. Receiving a visit is not the same as a conversion. The chain needs to be traced before the investment is justified.
What — GEO optimisation in practice currently means: be crawlable, be cited as authoritative by sources the AI trusts, have clear structured content that an AI system can extract and summarise accurately. This is recognisable. It is SEO. The fundamentals have not changed. The frame around them has been renamed.
The trade-off nobody mentions — to be cited in AI Mode or AI Overviews you must be open to Google's crawlers. The more you optimise your content for clean AI extraction — clear structure, direct answers, well-organised information — the better the AI's answer becomes, and the less reason the user has to click through to your site. You are optimising to be used without being visited. For some businesses that is a worthwhile trade — brand awareness, top-of-funnel presence, authority signalling. For most small businesses where traffic was the point, it deserves more scrutiny than it is currently receiving.
The infrastructure nobody is talking about
Brave Search processes 1.6 billion queries a month. Its market share as an AI chatbot is 0.1% to 0.6%. By that measure it barely registers. By a different measure it is one of the most significant players in the AI search ecosystem — because the Brave Search API supplies most of the top-ten LLMs with real-time web search data. For some of those LLMs Brave is the only search index supporting their AI answers. That includes Claude. When Claude answers a question that requires current web information it is frequently pulling from Brave's index, not Google's.
Microsoft closed Bing's public API in 2025. Google tightened scraping restrictions and initiated legal action against SerpApi. Brave — which built and maintains its own independent index of 40 billion pages — stepped into the gap. It is now the infrastructure layer underneath the logos people are debating. The ambulance everyone is chasing has Brave's engine in it.
This makes the "optimise for Claude" or "optimise for Gemini" framing even less useful than it already was. The content that appears in Claude's answers when it searches the web is content that is indexed by Brave. The content that appears in Gemini's answers is content that Google crawls. The content that appears in Perplexity's answers comes from a combination of sources including Brave's index. The question is not which AI logo to optimise for. The question is whether your content is in the indices these systems read — and whether it is structured, authoritative and accurate enough to be retrieved and cited when it is.
SEO ambulance chasing
The pattern since 2023 has been consistent:
A new AI platform arrives or gains coverage. The industry declares it the end of whatever came before. Agencies rebrand around it. Consultants produce urgent guidance. LinkedIn fills with posts explaining why you need to optimise for it immediately before your competitors do. The platform that was being optimised for last quarter quietly loses share. The cycle repeats.
2023 — optimise for ChatGPT.
2024 — optimise for AI Overviews and GEO.
2025 — optimise for Perplexity.
2026 — optimise for Claude. Optimise for Gemini.
2027 — whatever comes next.
The platform you were told to prioritise in March 2023 now has 46% share. The one you were told to ignore is at 27% — powered mostly by Google putting it in front of people who did not ask for it. The one nobody mentioned is at 10% and growing faster than any other major platform, with revenue that has now surpassed the one you started with.
If you had built your strategy around platform-specific optimisation you would have rebuilt it three times already. The sites that performed throughout this period are the ones that kept their technical foundations clean, their content accurate and specific, their crawlability intact. Because every one of these AI systems — ChatGPT, Gemini, Claude, Perplexity — reads the same underlying web. They all benefit from the same fundamentals. None of them are helped by a site that cannot be crawled, content that is vague and generic, or pages that answer the wrong question.
What to actually do
Optimising for AI/GEO/EI EI O. Now there's a question. Well you need a website that can be easily crawled, no barriers to entry (JavaScript), has a logical architecture, well linked to the important and relevant URLs and has words that actually mean something — not a phone book but a brochure or a novel. If you're not engaged with the conversation that's happening you'll end up with a picture of you in "the nip, with a dog."
Check your server logs, not just your GA4 dashboard. AI crawlers — OAI-SearchBot, Claude-SearchBot, PerplexityBot — do not execute JavaScript and are invisible to client-side analytics. If those user agents are not appearing in your logs, those systems are not visiting your site and no amount of GEO optimisation will change that until the crawlability problem is addressed first.
Before commissioning a GEO strategy, ask what you are trading. Clean extractable content helps AI systems give better answers. Better AI answers reduce the need to click through. If your business depends on visits that was the trade. Know it before you make it.
And when the next logo arrives — and it will — come back to the infrastructure that has not changed and will not change regardless of which AI system is winning this quarter:
- Architecture — clean site structure, logical hierarchy, no orphaned pages, crawl budget not wasted on things that do not matter
- Crawlability — robots.txt clean, sitemap submitted, no accidental noindex, internal linking logical, server logs checked
- Relevant content — specific not generic, accurate, aligned with what the page is actually for and what the user actually needs
- Speed — Core Web Vitals, image optimisation, no layout shift, pages that load before the user gives up
- UX — clear navigation, obvious next steps, a journey that makes sense to a person and to a crawler
These are not new. They were not new when Google launched. They will not be new when whatever replaces Gemini and Claude is announced with appropriate fanfare in 2027. The ambulance will still have Brave's engine in it. The infrastructure will still be the infrastructure.