AI search · 9 min read
Will AI replace SEO? Some of it, unevenly, and not the part you expect
The word SEO covers two different things: the work of being findable by a retrieval system, and one commercial deliverable — a click from a results page. Assistants depend entirely on the first and are visibly eating into the second. Treating those as one question is why the debate never settles.
Written by Zubair Afzal, FounderUpdated
The argument
The claim collapses on one dependency
Ask will AI replace SEO and you get two answers, both delivered with confidence, because the question hides two different things behind one three-letter word. One of them is a substrate: the work of being fetchable, parseable, correct and credible enough that a machine can find you and use you. The other is a deliverable: a person clicking a blue link on a results page and arriving on your site.
The substrate is not replaceable by the systems people think are replacing it, for a reason that is mechanical rather than rhetorical. An assistant answering a question about the present world does not know the answer from memory. It retrieves — from a search index, from a live fetch of the page, or both — and composes from what comes back. The pool it retrieves from is exactly the pool that search work exists to place you in. A technology cannot make its own input obsolete.
The deliverable is a different story, and this is where the people saying the field is finished have hold of something real. When a question is fully answered above the results, some of the people who would have clicked no longer need to. That does not affect all searches equally, and the unevenness is the whole practical point: it lands hardest on exactly the kind of page that a lot of organic programmes were built out of.
So the useful question is not whether the discipline survives. It is which parts of your particular programme were funded by a click that a machine can now supply for free. That is an audit, not a prediction, and you can do most of it yourself.
Exposure
The same programme, split into what is at risk and what is not
Run your own organic programme through this split before deciding anything about budget. The line does not fall between industries — it falls between page types.
Exposed to answer engines
- Definition and how-to pages built for a click
- Volume-led content with no purchase behind it
- Reporting that treats sessions as the outcome
- Forecasts based on informational search volume
Unaffected, or now worth more
- Crawl access, rendering and index inclusion
- Entity clarity across sources you do not own
- Original data nobody else can supply
- Commercial, local and branded queries
What would have to be true for this to be a genuine replacement
Rather than argue about the conclusion, set out the conditions. Each row is something that would have to hold for the strong version of the claim to be right.
| Dimension | What would have to be true | Where that actually stands |
|---|---|---|
| Assistants stop needing an index | Answers about current events, prices, availability and local businesses would have to come from the model itself rather than from a retrieval step. | They do not. Products that answer about the present world fetch or search, which is why they cite pages and why blocking their crawlers removes you from consideration. |
| Being findable stops mattering | A source that is unreachable, slow, or rendered only in the browser would have to be usable anyway. | The opposite. Retrieval is a harder filter than ranking was, because there is no page two to be found on. |
| Purchase-stage searches disappear | People would have to stop typing supplier names, product models, prices, locations and "near me" into a search box. | Those queries have the clearest commercial value and the least benefit from being summarised. A summary cannot take a booking. |
| Transactions move into the assistant | Buying, booking and enquiring would have to complete inside the assistant rather than on a business’s own site. | A live question rather than a settled one. Some of this is being built. It would change where the conversion happens, not whether being a known, described entity matters. |
| Brand and direct demand stop responding to visibility | Being named at the research stage would have to have no effect on who gets contacted later. | No evidence for that, and the mechanism runs the other way: shortlists are formed before the search that produces a click. |
| Informational clicks were the point | The commercial value of organic search would have to have been concentrated in question-shaped queries. | This is the one row where the sceptics are right about the risk and wrong about the size. Those clicks are the exposed ones. They were rarely where the revenue was. |
Underneath the question
Four different questions get asked in these words
When somebody asks whether AI has finished off search, they usually mean one of these four things. They have different answers, and answering the wrong one is why the conversation goes in circles.
- "Will Google send us less traffic than it does now?"
- Probably, on informational queries, and possibly already. This is the answerable version and it is measurable from your own data rather than from anybody’s forecast. Segment by query shape and page template, compare impressions against clicks, and look at where the two curves separate. If they separate on questions and hold on product and location terms, you have the common pattern.
- "Should we cut the budget?"
- Cut the part that was buying informational visits and keep the part that keeps you retrievable and coherent. The mistake we see most often is the reverse: businesses cancel the technical and entity work because it is boring and has no chart, and keep publishing top-of-funnel articles because there is a content calendar and somebody is already writing them.
- "Will the job still exist in five years?"
- The job has already changed name three times and survived each one. What is genuinely disappearing is the version of it built on rank reports and monthly session graphs, because the reporting no longer describes the outcome. That is a change to the craft rather than a removal of it, and it is not a comfortable change for people whose value was the report.
- "Are we being sold a new thing to replace the old thing?"
- Frequently, yes, and the tell is easy to check. Ask what a proposed AI search line item does that the organic line item does not. If the answer is a list of things you were already paying for — crawlability, structure, accurate copy, credible authorship — you are being charged twice for one job. There is real new work in this category, but it is narrower than the invoices suggest.
Where we stand
What we are willing to say about this, and what we are not
What we will say
- Informational content aimed at capturing a click is a weaker investment than it was, and we will tell a client to stop commissioning it.
- Retrieval, entity clarity and purchase-stage coverage are more important now than before, not less.
- The exposure varies enormously by page type, and yours can be measured from your own Search Console data.
- Some of what is sold as new work under an AI heading is organic search work with a different label on the invoice.
What we will not say
- That we know what search traffic will look like in three years. Nobody does, including the people publishing charts of it.
- That our own services are somehow unaffected by a change we are describing as significant.
- That being included in an AI answer can be promised, arranged, or bought.
- That doing nothing is safe because the fundamentals never change. Something real is moving here.
The evidence, and the work either way
This page is the argument. These are the numbers behind it and the practical consequences.
Questions
The questions this comes with
Is SEO dead?
No, but one thing people bought under that name is dying, and pretending otherwise is how agencies lose clients. The purely informational page that existed to catch a definitional search and convert one visitor in four hundred is a worse investment now than it was three years ago.
The underlying work — being retrievable, being correct, being the clearest source on a claim, being chosen for a purchase-stage query — is not only intact but harder to substitute, because every assistant depends on it.
Should we cut our SEO budget?
Cut the part of it that was buying informational clicks and move that money down the funnel. Do not cut the part that keeps the site retrievable and the entity coherent, because that is the precondition for every surface, old and new.
The honest version of this exercise is uncomfortable: list your organic landing pages by revenue contribution rather than by sessions, and see how much of the programme survives the sort. Most programmes have a long tail nobody has ever justified commercially.
Will people stop visiting websites entirely?
There is no evidence for that and a lot of mechanism against it. Assistants transact very little, and the things people buy from a business — a booking, a quote, a demo, a checkout — still happen on a site the business controls.
What is changing is the share of visits that arrive to answer a question rather than to do something. A site that only ever answered questions is more exposed than one that also completes a transaction.
If we appear in an answer without a click, is that worth anything?
Sometimes, and it is worth being precise about when. Being named accurately in an answer at the shortlisting stage affects which three companies get looked up later, and that shows up as branded and direct traffic that nothing in the analytics attributes correctly.
Being cited in an answer to a question that was never going to produce a customer is worth close to nothing, and it is the kind of thing that gets put in reports because it is easy to screenshot.
Do we need a separate AI search retainer?
Usually not a separate one. Most of what is sold under that heading is retrieval and content work that belongs in an organic programme anyway, and splitting it into two invoices mainly benefits the supplier.
A separate scope makes sense in one case: when a business needs its entity description corrected across sources it does not own, which is genuinely different work from publishing pages and often needs different people.
How would we know if this prediction turns out to be wrong?
Watch the ratio rather than the total. If clicks fall while impressions hold on question-shaped queries, the shift is happening to you. If commercial and branded queries start behaving the same way, the argument on this page is wrong and the exposure is much broader than described.
We would rather be told that than defend the position. The measurement is available to anyone with Search Console and an afternoon.
Find out how exposed your own programme is
Give us read access to Search Console and we will show you which of your organic landing pages are in the exposed group and which are not. It is a short piece of work and the answer is usually not what people expect in either direction.
Related
Where to go next
Last updated