How AI search changed SEO — and what still works
Every SEO conversation in the last two years eventually turns into a panic about AI Overviews and ChatGPT. Some of that panic is earned. A lot of it is not. Here is what actually changed, what did not, and where that leaves the work you do every day.
5-day free trial · 30-day money-back · price locked for life
Search has not been replaced. It has split. Google still processes billions of queries a day, and most of them still return a page of blue links that people click through. What changed is that a growing slice of queries now get answered directly, inside an AI Overview at the top of Google, or inside a chat window in ChatGPT, Perplexity, Claude, Gemini, Grok or DeepSeek, with no click required at all. For those queries, the win condition is no longer a top-three ranking. It is being one of the two or three sources the model names in its answer.
That split is why the same page can be doing well in Search Console and invisible in AI answers, or the reverse. They are graded differently, by different mechanisms, and improving one does not automatically improve the other.
Three things that are genuinely different now
1. The answer can end the session before the click happens
An AI Overview or a chat answer can fully satisfy a query without the reader ever visiting a site. For informational queries especially, this means a page can lose most of its organic traffic even while it keeps its ranking, because the ranking no longer guarantees a visit. Traffic decline on a page that still ranks well is now a real, distinct failure mode that keyword-position tracking alone will not catch.
2. There are now several audiences with different crawlers and different rules
Googlebot indexes for classic search and, separately, Google-Extended feeds Google’s own AI features. GPTBot crawls for ChatGPT, PerplexityBot for Perplexity, ClaudeBot for Claude, and each respects robots.txt independently. A site can be wide open to Googlebot and quietly blocking every AI crawler, or the other way round, without anyone noticing until citations (or rankings) stop showing up.
3. Retrieval, not just ranking, decides who gets seen
Classic search ranks a fixed index against a query. Generative engines run a retrieval step first, pulling in a handful of candidate pages, and then the model decides which of those candidates to trust and quote. A page can be well-optimized for keywords and still never get pulled into that candidate set if it does not state its facts in a form the retrieval step recognizes as directly relevant.
The fundamentals that still hold
It is easy to read the last section and conclude that SEO fundamentals stopped mattering. They did not. AI engines still need a working, crawlable, fast, well-structured site to pull information from in the first place. None of the following got less important:
- Technical health. A page that 404s, loads slowly, or is blocked by a stray robots.txt rule cannot be cited by anything, human or model.
- Original research and first-party data. A model looking for a quotable, specific claim favors a page that actually ran the numbers over one that paraphrases someone else’s.
- Internal linking and site structure. Both Googlebot and AI crawlers rely on links to discover and understand how pages relate to each other.
- E-E-A-T signals. Author bios, credentials and a clear “who wrote this and why should I trust them” remain a trust signal for classic ranking and a citation signal for AI retrieval.
- Backlinks. They still carry authority signal into both systems, even if they are no longer the single dominant one.
In practice, most of the SEO work that was worth doing in 2022 is still worth doing in 2026. The difference is what you add on top of it, not what you throw out.
Assumptions that stopped holding
A handful of assumptions that used to be safe defaults are no longer safe:
- “If I rank, I get the click.” Not for queries an AI Overview can fully answer. Ranking and traffic are now two separate metrics to track.
- “One robots.txt policy covers search.” It now has to name AI crawlers explicitly, or you are guessing whether they are allowed in.
- “Long, comprehensive articles always win.” Length still helps depth and topical coverage, but a 3,000-word article that buries its key facts in paragraph six is harder for a model to extract a citable answer from than a shorter page that states them plainly near the top.
- “Keyword rank trackers tell the whole story.” They report position in the blue links only. They say nothing about whether your site was named inside an AI answer for the same query.
SEO then vs. SEO now
| Before | Now |
|---|---|
| Rank for a keyword, win the click | Rank for a keyword and get quoted inside the AI answer |
| One result set: Google’s ten blue links | Multiple result sets: blue links, AI Overview, and a separate answer inside ChatGPT, Perplexity, Gemini |
| Success measured in clicks | Success measured in clicks plus citations and share of voice inside AI answers |
| Content optimized for one crawler (Googlebot) | Content optimized for Googlebot and half a dozen AI crawlers with different rules |
| Backlinks were the dominant trust signal | Backlinks still matter, but clear entities and quotable facts carry more weight in retrieval |
A practical way to prioritize
You do not need to rebuild your entire content operation. Keep the technical, structural and authority work you already do, since it still underpins everything, and layer on a second pass aimed specifically at AI retrieval: check whether AI crawlers can reach your site, add or refresh an llms.txt file, rewrite the first two or three sentences of your key pages so they state the answer plainly, and add structured data where you already have the underlying facts in prose. For the full breakdown of how the two disciplines differ and a five-item action list, see our guide to SEO vs GEO.
This is exactly the split Climb AI is built around. The SEO Agent keeps handling keyword research, content writing, technical audits and publishing, the work that still wins the ten blue links. The GEO Agent runs the second pass on top of it, scoring how citable your existing pages already are and generating the fix files, llms.txt, schema, FAQ sections, author bios, that AI engines look for. Both are included in every plan, because most sites need both audiences covered, not a choice between them.
Three ways out, before you are ever stuck.
Both agents, all 13 modules, from the first minute. Cancel inside the trial and you are never charged.
If the first month does not earn its place, email us and we refund the charge in full. No forms, no retention call.
When we raise prices for new customers, yours does not move for as long as you stay subscribed.
Stop doing SEO. Start having it done.
Both agents, every module, from $49.99/month. Cancel in one click.
5-day free trial · 30-day money-back · price locked for life