AI Search in Mid-2026: A Big Update on SEO, AEO and GEO

Six months ago I wrote about using AI to do your SEO. Halfway through 2026, the bigger story is the reverse: doing SEO for AI. The acronyms have multiplied (SEO, AEO, GEO, LLMO, "search everywhere optimisation") and underneath the jargon, the way people find websites has genuinely changed. This is my big mid-year update: what the data actually says, what's changed in the last few months, and the practical checklist I'm applying to this very site.

The acronyms, in plain English

Let's get the definitions out of the way, because half the confusion in this space is people using three names for the same thing.

• SEO (Search Engine Optimisation) is the classic discipline: ranking in the blue links on Google and Bing. Still alive, still the biggest single source of traffic for most sites.
• AEO (Answer Engine Optimisation) means optimising to be the answer: featured snippets, voice assistants, and the AI answer boxes that sit above the blue links.
• GEO (Generative Engine Optimisation) means optimising to be cited inside generated responses from ChatGPT, Perplexity, Claude, Gemini and Google's AI Mode. You're not chasing a rank; you're chasing a mention.

Here's the part I find most telling: Google itself now publicly frames preparing for generative search as part of SEO, not a separate dark art. The techniques overlap heavily. Think of AEO and GEO as new distribution channels for the same underlying asset (clear, structured, trustworthy content) rather than three separate jobs.

The numbers that changed my mind

I was fairly relaxed about all this a year ago. The mid-2026 data made me sit up. A few figures from recent industry studies (all vendor research, so treat the exact numbers as directional, but the direction is unmistakable):

• AI Overviews now appear on roughly half of Google searches and are in front of over 2 billion monthly users. When an AI Overview shows up, organic click-through rates drop by around 60%.
• Google's AI Mode is even more aggressive: around 93% of AI Mode sessions end without a single click to a website.
• Publisher traffic is genuinely falling. Google search traffic to publishers dropped about a third in the year to late 2025, and smaller sites are being hit hardest.
• Ranking well no longer guarantees being cited. The overlap between Google's top 10 organic results and the sources AI engines actually cite has collapsed, from roughly 70-75% in mid-2025 to somewhere between 17% and 38% in early 2026, depending on whose study you read. AI engines are developing their own taste in sources.
• The engines behave very differently. Perplexity cites sources on about 97% of answers; ChatGPT cites on closer to 16%. And the market itself is moving: recent traffic reports show Claude's share of AI search growing sharply through 2026 while ChatGPT's slips.
• Perspective check: AI referral traffic grew over 1,000% last year… and still adds up to less than 1% of total web traffic. Gartner expects AI assistants to handle about a quarter of searches this year and half by 2028.

Put together, that's the story of 2026 so far: traditional search still delivers most of the visitors, but the growth, and increasingly the influence, is in the answer layer. If an AI can't read, verify and cite you, you're invisible in the place where a growing share of buying decisions start.

What actually gets you cited

The good news: getting cited by AI engines is not magic, and most of it is stuff good SEO practitioners already do. Based on the citation studies published this year, five things consistently move the needle.

1. Let the bots in

This is step zero and it's shocking how many sites fail it. Check your robots.txt and make sure you're not blocking GPTBot, OAI-SearchBot, PerplexityBot, Perplexity-User, ClaudeBot or Google-Extended. If a crawler is blocked, no amount of content quality matters for that engine: you simply don't exist to it. (Blocking them is a legitimate choice for some publishers, but make it a choice, not an accident.)

2. Answer-first writing

Start every section, ideally every page, with a direct, self-contained answer in the first sentence or two, then elaborate. LLMs lift these passages as canonical answers, and studies this year suggest answer-first formatting raises citation likelihood by around 40%. It's the single highest-leverage content change, and conveniently it also makes your writing better for humans.

3. Structure and schema

Convert dense paragraphs into tables, numbered lists and FAQs where it makes sense. Add schema markup: pages with proper structured data see citation rates several times higher in some studies. FAQ blocks appear to carry the strongest individual signal. Keep the question text identical to your visible H2/H3 headings and the answers tight, around 40-60 words.

4. Freshness that machines can verify

AI engines strongly prefer recent content. In one analysis, around half of Perplexity's 2026 citations pointed at pages published or updated in 2025. Make sure every page carries an accurate dateModified in its schema, and actually revisit your cornerstone posts instead of leaving them to rot.

5. Be quotable, and be consistent everywhere

Engines love specific, quotable statements (statistics, concrete claims, named entities) over vague marketing prose. And they cross-check: your site, your social profiles, directories and third-party mentions should all tell the same story about who you are and what you do. Off-site presence in reviews, listicles and communities feeds the models too; those reputational signals just take longer (think 30-90 days) to show up than on-site fixes, which can surface in Perplexity within days and in ChatGPT within a few weeks.

One honest caveat

The GEO industry has a gold-rush feel right now: new tracking tools, agencies and "AI visibility platforms" launching weekly, each with a study proving their approach works. Some of it is genuinely useful; plenty is repackaged SEO with a 3x price tag. My rule of thumb: if a tactic wouldn't also improve your site for a human reader or a traditional crawler, be sceptical. Everything in the checklist above passes that test.

The bottom line

SEO isn't dead; it's growing an answer layer on top. The blue links still pay the bills in mid-2026, but the collapse in overlap between what ranks and what gets cited means you can no longer assume good rankings buy you AI visibility for free. The fix isn't a new discipline. It's a sharper version of the old one: let the crawlers in, answer questions directly, structure everything, keep it fresh, and be consistently, verifiably yourself across the web.

I'm working through this checklist on this site over the coming weeks (llms.txt, schema audit, answer-first rewrites of the top posts) and I'll share the before/after citation data once there's enough of it to be meaningful. If you're doing the same on your site, I'd genuinely love to compare notes.