Is SEO dead? What changes with AI answers, and what does not
No, and the honest version of the answer is more useful than the slogan. Assistants retrieve from a web index, cite pages that answer a question directly, and take some clicks that used to land on those pages. The mechanics underneath are unchanged: a crawler finds your page, an index keeps it, something ranks it, and a person decides whether to click. What moves is where the decision happens.
By Théophile Louvart, founder of Porteur · Updated 15 September 2026 · Markdown
The short answer
- Search engines still crawl, index and rank. Nothing about that changed.
- Assistants retrieve from a web index and cite pages. To be cited you have to be in the index and answer the question.
- Some informational queries send fewer clicks, because the answer now fits in a box above the results.
- Commercial, comparison and brand queries still send the click, because the searcher wants your page, not a summary.
- The new prize is a citation. There is no way to submit for one, and no way to buy one.
So the work did not die. Part of it moved from earning a position to earning a mention, and the two are won by the same page.
What actually changed
| Then | Now |
|---|---|
| One results page, ten links, you wanted one of them | A results page that may open with a written answer citing a handful of pages, plus the ten links |
| One engine to be found in | Google and Bing, plus assistants that retrieve from those indexes or their own crawl |
| Rank tracking told you roughly where you stood | Rank tracking still works for the links; nothing reports whether an assistant named you |
| A page ranked and got clicks | A page can rank, be quoted, and get fewer clicks than it did |
| Crawlers were Googlebot and Bingbot | Also OAI-SearchBot, Claude-SearchBot, PerplexityBot and the training crawlers, each with its own robots token |
The last row is the one founders miss. Blocking the answering crawlers in robots.txt makes a site invisible to assistants, and blocking a training crawler does not remove what was already collected.
What did not change
Google still works in three stages: crawling, indexing and serving. An assistant that searches sits on top of an index built the same way. Everything that made a page findable still makes it findable.
- A page has to be crawlable and indexable. A noindex shipped from staging is still the most expensive accident in this business.
- A page has to answer the query. Assistants cite pages that answer directly, which is the same thing ranking rewards.
- Links still count as one signal among many, and they still bring the visitors that turn into brand searches.
- Brand demand still beats everything. Queries for your own name convert best and no assistant intercepts them.
- Speed, structure and clarity still decide whether a page can be read by a crawler that does not run JavaScript, which is most of the answering ones.
What is measurable, and what is guesswork
| Measurable | Not measurable |
|---|---|
| Impressions, clicks and average position per query, from Search Console | How many clicks an AI overview took from you: those clicks are reported inside the same web search totals |
| Whether clicks fell while impressions held | Whether the fall was the overview, a new rival, or your own title change, without checking each query |
| Whether an assistant names you today, by asking it | How often it names you across everyone's conversations: nobody publishes that |
| Visits arriving with an assistant's referrer in your analytics | The buyer who asked an assistant, read the answer and typed your name in later |
Anyone quoting a precise figure for your site's losses to AI answers is modelling. The defensible measurements are the ones in the left column, plus a list of buyer questions you ask the assistants every month and log.
Which queries lose, and which do not
This is the part that decides where a small product should spend its next quarter.
- Losing clicks: definitions, simple how-tos, anything answered in three sentences. The box is enough.
- Holding: comparisons, alternatives, best-of and pricing queries. People check the detail, and they want to see the thing itself.
- Untouched: your brand, your docs, your login, and queries where someone wants to do a task rather than read about it.
- Growing in value: the pages assistants cite when they recommend a category, which are the lists, comparisons and review pages your product appears on.
A library of definition pages was always the weakest asset a product site could build. That is more true now, and the fix is the same as it always was: write the pages a buyer needs, not the pages a keyword tool suggested.
Five things to do differently this year
Check your robots.txt lets the answering crawlers in
OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot and Googlebot. Decide separately about the training crawlers; that is a different question with different trade-offs.
Move the answer to the top of every page
The first two sentences answer the query under a heading that matches it. This wins featured snippets, citations and impatient readers at the same time.
Put the facts in quotable shapes
Lists, tables, a figure with its unit, consistent product naming. An assistant quotes what it can lift cleanly.
Get onto the lists that answers already cite
Category roundups, comparison pages, directories and review sites. Being named there is the most reliable route into an answer, because there is no submission mechanism for the answer itself.
Track citations by hand, monthly
Ten buyer questions, asked to each assistant, with who was named written into a sheet. Twenty minutes a month, and it is the only honest measurement available.
What to ignore
- Predictions that search is finished. The same claim has been made about every change since the first results page, and the index is still the thing everything reads from.
- Tools promising to submit your site to AI assistants. There is no submission mechanism.
- Precise percentages of traffic lost to AI, for your site. They are estimates with unknown error.
- Advice to block Googlebot to stay out of overviews. There is no separate opt-out, and blocking removes you from Search too.
- Rewriting your whole library for one bad month before checking the dates against Google's status dashboard and your own deploys.
The founders who do well in this period are doing the same things they were doing in 2019, with the answer higher up the page and a monthly check on who the assistants name.
Questions
No. Assistants retrieve from a web index, so a page still has to be crawlable, indexed and good enough to be chosen. What changed is that some informational queries send fewer clicks, and that being cited in an answer is now worth something on top of ranking.
It replaces some of the typing. The judgement it cannot do is choosing which pages deserve to exist, what claim each makes, and which facts go on them, which is most of the work that produces traffic.
Not instead. The assistants retrieve from Bing's index, Google's, or their own crawl, so being findable is the precondition for both. The extra work for assistants is structural: answer early, use lists and tables, name things consistently and get onto the lists they already cite.
Ask it. There is no report, so the method is to keep a list of the questions your buyers ask, put them to each assistant on a schedule, and write down who gets named and which pages get cited.
It has said the clicks it sends are of higher quality and it publishes no separate figures. Overview clicks are reported inside web search totals in Search Console, so neither Google nor you can split them out for your site.
Check that the answering crawlers are allowed, then rewrite the first two sentences of your best pages so they answer the query directly. Those two changes cost an afternoon and affect ranking, snippets and citations at once.
Sources
Check my site, free
Paste your URL and the free check reads your site, the searches around it and who holds them, including which pages an assistant names in your category, in about thirty seconds.
- Free check, no card
- Read-only, your own accounts
- Readable by your agent
Read next
- GuideGEO vs SEO: what changes, what does not
- GuideGenerative engine optimization (GEO): how to be named by AI answers
- GuideWhat AI Overviews do to your traffic, and how to tell
- GuideHow Google Search works: crawling, indexing, ranking, in plain words
- GuideWriting pages AI assistants can cite
- GuideWhat SEO is in 2026, explained to a product founder
- Guiderobots.txt for AI crawlers: GPTBot, ClaudeBot, PerplexityBot and what to allow
- GuideHTTP/2 and HTTP/3: what they change for a site’s speed