AI SEO: what actually changed in 2026.
'AI SEO' is being sold as a new discipline, a dead discipline and a rebrand of the old one, depending on who is selling. From inside client accounts, here is what has observably changed in the last eighteen months, what has not, and what we now do differently.
What changed
Fewer clicks for the same rankings
Google AI Overviews and AI Mode answer more informational queries on the results page, and click-through from the classic listings underneath has fallen for those queries. On our own site the pattern is stark: the informational articles earned impressions and almost no clicks this year, while commercial and local terms, where the user still wants a supplier rather than an answer, kept converting. Client accounts we have checked show the same shape.
Buyers asking assistants before they search
More of the B2B and healthcare enquiries we take now mention having asked ChatGPT or Gemini for a shortlist first. The prospect's first Google search is your brand name, because an assistant already suggested you. If the assistant did not, you were never in the running.
Demand shifting inside the keyword data
Our own search volume tracking for agency terms shows generic queries such as 'digital marketing agency' plus a city falling 20 to 45 percent year on year, while specific queries ('ai seo', 'google ads management', 'local seo services') grew. People are asking narrower questions, and many of the broad ones are going to assistants instead.
Crawler policy as a business decision
Sites now choose which AI crawlers to allow. Blocking everything protects content and removes you from AI answers. Allowing everything feeds training sets with no benefit. The sensible middle is to allow the crawlers that cite sources and block the ones that only harvest; that is our own robots.txt policy.
What did not change
- Google's index still underpins most AI answers. If you are not retrievable, you are not citable.
- Authority still decides competitive terms. Links from relevant, credible sites remain the ceiling on results.
- Commercial and local intent still converts through classic results and the map pack.
- Thin, mass-produced content still loses, and AI made it cheaper to produce, so it loses faster.
What stopped working
- Publishing volume for its own sake. A hundred generic articles now earn less than ten specific ones. We measured this on our own site and rebuilt the content programme as a result.
- Programmatic city pages with swapped place names. We retired ours this month and redirected them to real pages.
- Keyword-stuffed titles. Searchers' phrasing still matters, but the page has to answer, not just match.
- Reporting rankings alone. A position-three ranking on an AI-answered query can produce zero clicks.
The four things to change first
- Answer-first pages. State the answer in the first hundred words, use the question as the heading, put numbers and definitions in plain text. Explained in what is generative engine optimisation.
- Entities and evidence. Named authors, an organisation page a machine can verify, case studies with clients and numbers, consistent details across the web.
- Measure presence, not just position. Track whether AI engines mention and cite you for your buyer questions, alongside organic revenue. We built SearchInsights AI for this.
- Fix the crawl and the manifest. Allow the right crawlers, publish an llms.txt, keep schema accurate.
What it means for budgets
Less on content volume, more on authority and on the small number of pages that carry commercial intent. Less on rank reports, more on reconciling search to enquiries. For most Australian businesses the total does not need to go up; it needs to move. Our AI SEO agency page sets out how we run the programme, and the GEO service covers the AI-answer side in detail.
Want to know whether AI engines mention your brand at all today? The free SEO audit includes a read on your AI visibility.
Get the free SEO audit