Keyword ranking -
I’ve now sat through three separate “keywords are dead” announcements. Hummingbird was the first one, then the voice search panic, and now this one, with AI. Each time there’s a confident post, a few nervous meetings, and then everyone quietly carries on typing keywords into the same tools as before.
So I’m a bit allergic to the claim. But I’ll be fair to it, because this time something really has shifted, and it’s worth knowing what.
Why people say it
The old way of doing SEO had it coming, to be fair. Pick a phrase, cram it into the title, the headings and the intro, then sprinkle it every few paragraphs like salt, and go buy links until Google caved. The pages were awful to read. Everyone knew it and did it anyway because it worked.
AI tools don’t fall for that. A language model reads the whole page and works out what it’s about. Say “cheap flights” fifteen times without telling anybody how to actually find one, and the model isn’t fooled. Neither, to be fair, is Google these days.
So yes, the repeat-the-phrase trick is done. If that’s what people mean by “keywords are dead”, I’ll happily agree.
But the words still matter
People still have to ask for things somehow, and the words they pick are what get matched to your page.
Google first. BrightEdge puts organic search at about 53% of trackable website traffic, and plenty of that begins with somebody typing three or four words and pressing enter. Call those words whatever you like. They’re keywords.
The AI tools are a funnier case. Ask ChatGPT or Perplexity something and, behind the scenes, it often fires off a few searches of its own before it writes a word back to you. Whichever pages match those searches best get pulled in. Which means the wording on your page still has a say in whether you show up at all. You just can’t get away with one phrase on repeat anymore. You have to cover the subject properly.
Take an accounting software company. Ten skinny pages, each aimed at a slightly different phrase, was the usual plan once. These days I’d put that effort into a single decent one: how invoicing works, where small firms go wrong, what to check before you switch. It’ll often beat all ten.
The click problem
This is the bit that actually stings, and it’s backed by numbers.
Pew Research tracked real browsing across nearly 69,000 Google searches. When an AI summary appeared, people clicked a normal result about 8% of the time. With no summary, 15%. Ahrefs looked at it from the ranking side and found the page in first position lost about a third of its click-through rate when an AI Overview sat above it.
First place, a third fewer clicks. I’d be grumpy about that too.
There’s a flip side, though. Semrush found that visitors coming from AI search convert at a rate making each one worth roughly 4.4 times a normal organic visitor. They’ve read an explanation, then bothered to click through anyway. Those people are nearly sold.
So fewer clicks, and better ones. Anybody telling you only one half of that is either panicking or selling comfort.
Does the research back any of this up?
Some of it. Back in 2023 a group of academics built a benchmark called GEO-Bench and threw a whole bunch of tactics at AI-generated answers to see what moved. Some boosted a source’s visibility by up to 40%. The ones that helped were things like adding quotations, statistics and proper citations. Boring stuff. An editor would have asked for it anyway.
What rarely gets mentioned is that the results varied a lot from one subject area to another. Something that worked brilliantly in one field did almost nothing in the next. I’d take that as a reason to distrust any universal checklist, including ones from people I like.
And citations wobble. Studies suggest 40 to 60% of the sources cited in AI answers change from one month to the next. Getting named once is nice. It doesn’t last unless you keep at it.
What I’d actually do
Still keyword research, yes, but I’d treat what comes out of it differently. Instead of a sheet of phrases to repeat, I’d want a list of the questions customers really ask, grouped by subject, with a rough idea of how close to buying each person is.
Then I’d put the answer high on the page. AI tools lift small chunks and ignore the rest, so if your real answer is hiding in paragraph nine, don’t expect it to get quoted.
I’d also pay more attention to what other people say about the business. Reddit, reviews, LinkedIn, YouTube. The tools read all of that, and it shapes what they end up saying about you. A site that’s brilliant but never gets mentioned anywhere else is in a slightly odd position now.
Picking a provider
If you’re looking at SEO services at the moment, this whole argument is a decent filter.
Run from anyone who says it’s all over and sells you something brand new with a fancy acronym. Run just as fast from the one whose entire pitch is a list of phrases and a promise to rank for them. I’ve met both, and neither will do much for you.
The good ones sit somewhere in the middle. They still do keyword research. They talk about topics and questions, can show you where you’re being cited in AI answers, and tell you straight that results take months, not weeks.
So, dead or not?
Not dead. The mechanical version is, and good riddance to it. What remains is the thing it was always a clumsy stand-in for: working out what people want and giving them a decent answer. That was the job all along. It’s just harder to fake now.
Questions people ask me
Should I take keywords out of my content?
Nope. Write like a normal person, use the words your customers use, and make sure the page actually covers the subject. Counting how many times a phrase appears is a waste of your afternoon.
Do AI search engines use keywords at all?
Kind of, though not directly. The tool breaks your question into a few searches and hunts for pages that fit them, so your wording still gets weighed.
Is long-tail still worth going after?
More than before, I’d say. Specific questions are exactly what people ask AI tools, and the people asking them are often further along in deciding.
Will keyword tools stop being useful?
I doubt it. They’ll probably lean more towards questions and topics and less towards bare phrases. Don’t take the numbers as gospel, compare them with what real customers say to you.
How long until I see results?
A strong new page can get picked up within a few weeks. Anything that leans on your wider reputation tends to take six to twelve months, which is more or less how SEO has always worked. more
