Straight answer: Yes, keyword research is still important, but what you’re researching for has changed. You’re no longer just trying to rank on a results page, you’re trying to become the source that ChatGPT, Google’s AI Overviews, and Perplexity pull from when someone asks a question. The words haven’t stopped mattering. The prize you’re playing for has.
What changed and what didn’t
I’ve been doing this for over fifteen years, and I’ve watched three or four “SEO is dead” moments come and go. Mobile-first indexing was supposed to kill it. Voice search was supposed to kill it. Now it’s AI search doing the killing, apparently. And every time, the people saying it’s dead are half right and half lazy.
Here’s what’s true: fewer people are clicking through from search results because AI Overviews and chat tools answer the question right there on the page. Fewer people are typing three-word phrases into a search box because they’re having a conversation with an AI instead. That part is real and it’s not going away.
Here’s what hasn’t changed: search engines and AI models still need to understand what a piece of content is about before they can decide whether to show it, cite it, or quote it. That understanding still starts with words. Specific words, real phrases, the actual language your customer uses when they’re stuck or curious or annoyed. Keyword research was never really about stuffing a term into a page ten times. It was always about listening to how people describe their problem. That hasn’t stopped being useful, it’s arguably become more useful, because AI models are stitching together answers from multiple sources and the sources with the clearest, most specific language win the citation.
The client whose traffic dropped while her rankings stayed the same
I want to tell you about a client I worked with last year, a coach who ran a small consultancy helping people transition into project management roles. She had a blog post ranking position two for “how to become a project manager without experience.” Position two. Good traffic for years. Then, over about four months, her clicks on that exact page dropped by roughly 40 percent while her ranking position barely moved.
We checked Google Search Console and the pattern was obvious once we looked at impressions versus clicks. Impressions were flat or even slightly up. Clicks were down. Google’s AI Overview had started answering that exact question directly in the results page, summarising the steps, and most people never scrolled past it.
Now here’s the bit that surprised her. We didn’t abandon the keyword. We rewrote the page to answer a narrower, more specific version of the same question: “how to become a project manager without experience if you’re over 40,” because that was a phrase showing up in her impressions data that had almost no competition. Within seven weeks that page was getting quoted, word for word in places, inside AI Overview snippets. Her overall organic clicks across the two pages ended up higher than the original page alone had ever pulled in. The research didn’t stop mattering. It just had to go one layer deeper than “what term gets the most searches.”
The uncomfortable part nobody likes admitting
Here’s the bit that makes a lot of marketers uncomfortable: chasing high-volume keywords is now often a waste of your time, and it was already a questionable strategy before AI search made it worse. If a term gets 10,000 searches a month and it’s a broad, informational question, there’s a very good chance Google or ChatGPT is going to answer it directly and you’ll never see the click. High volume used to mean opportunity. Now high volume on a simple factual question often means you’re writing content for an AI model to summarise for free, not for a human to visit your site for.
The keywords that still convert into actual visits are the ones with a bit of friction in them, the ones that need a real opinion, a comparison, a personal account, a number that isn’t settled fact. “What is a keyword” gets answered in one AI sentence. “Should I hire an AI consultant or do this myself if I run a five-person agency” doesn’t get answered in one sentence, because it needs judgement, and that’s exactly the kind of query where a human still clicks through to read a real answer from a real person.
I’ll say the quiet part louder: a lot of the keyword research industry has spent two decades optimising for search volume as the main signal, and that habit is now actively hurting the businesses still doing it. Volume tells you how many people are asking. It tells you nothing about whether they need to leave the search results page to get the answer. In 2026, that second question matters more than the first.
What keyword research needs to do now
I don’t say this to be dramatic, but the job has split into two different tasks that used to be one.
- Task one: find out what language your audience uses, so you show up in the sources AI models draw from at all.
- Task two: find the specific, narrower version of that question where a human still wants to click through and read your full answer, not just glance at a summary.
Most keyword tools were built for task one and still are. A good keyword explorer tool will show you search volume, related terms, and difficulty scores, and that’s still useful groundwork. But you now need to layer a second filter on top: does this question have a clean, factual, one-line answer, or does it need context, opinion, comparison, or personal circumstance? If it’s the first, expect an AI Overview to eat your click. If it’s the second, you’re in business.
A five-step process I use with clients now
- Pull your existing top 50 pages by impressions from Search Console, not clicks. Impressions tell you what you’re already being shown for. Clicks tell you what’s surviving contact with AI Overviews.
- Sort by the biggest gap between impressions and clicks. Those are the pages where an AI summary is most likely eating your traffic. That gap is your signal, not your ranking position.
- For each of those pages, find one narrower, more specific angle within the same topic, something with a condition attached (“for beginners over 50,” “on a budget,” “if you’re self-employed,” “without a degree”). Specificity is what survives.
- Check search volume on the narrower version, but don’t reject it for being small. A term with 40 monthly searches and zero competition that converts beats a term with 5,000 searches that gets summarised away every time.
- Write the new or updated page with the answer stated plainly in the first two sentences, then back it up with a real example, number, or step below it. AI models tend to quote the clean summary near the top; humans tend to stay for the proof underneath.
That’s not theory. That’s the exact sequence we ran for the project management coach, and it’s what I’d tell any small business owner to do before spending another penny on content this quarter.
Where keyword research still wins outright
There are entire categories where keyword research is doing more work than ever, not less.
Local and transactional searches are one. If someone searches “AI consultant near me” or “how much does an AI consultant cost,” they’re not looking for a summary, they’re looking for options to compare and someone to contact. AI Overviews handle informational questions well. They handle “who should I hire” questions badly, because that decision needs judgement and local context that a generic summary can’t give. I wrote about this when I covered why AI matters for business, and the pattern holds across most service industries, not just mine.
Job and career searches are another. My piece on the broken “remote jobs near me” search covers this in more depth, but the short version is that people searching for work still want a list of real, current, specific listings, not a summary of what remote work generally means. Volume and specificity both still matter there, arguably more than they used to, because the low-effort searches have been swallowed by AI answers, leaving the high-intent ones as the only traffic left worth having.
And anything with a number attached, prices, timelines, pass rates, salary ranges, tends to keep pulling clicks because people don’t trust a single AI-generated figure for a decision that costs them money. They want to see the source, check the date, and often check a second source too. If your keyword research turns up a “how much does X cost” question, that page is usually still worth building, with real figures, not rounded guesses.
Where it stops helping
Let’s be blunt about the other side too, because pretending nothing has changed does nobody any favours. If your keyword research keeps pointing you toward broad definitional questions like “what is SEO” or “what does AI mean for marketing,” you are building content for a machine to summarise, not for a person to read. I’d rather see a client spend that hour building three narrow, specific pages than one broad one that Google will answer itself in the first three lines of the results page.
Tracking is where this gets uncomfortable too. A lot of businesses still watch their rank tracker as the main scoreboard, and I understand why, it’s the easiest number to check on a Monday morning. But rank position without a click-through comparison is now close to meaningless. You can hold position one on a term and still lose 40 percent of the traffic you used to get from it, exactly as happened to my client. If you’re only watching position, you’re watching the wrong number.
You don’t need expensive tools to start this
One thing I’ll say plainly: you don’t need a five-figure SEO software budget to do any of what I’ve described above. Search Console is free and it’s the single most honest source of data you have, because it shows you what you’re already being shown for, which is the real starting point. Beyond that, autocomplete, the “People also ask” boxes, and even asking ChatGPT what questions people commonly have about your topic will surface the specific, narrower phrases I keep talking about. I’ve laid out the exact free method, no paid tools at all, in my guide to doing keyword research without expensive tools, and it’s the same process I use for my own site when I’m not paying for a subscription that month.
If you take one thing from this post, take this: stop asking “is this keyword still relevant” and start asking “would a human still need to click through to get this answer.” That second question is the whole game now. Keyword research hasn’t died. It’s just had to grow a second job it never used to need.
Frequently asked questions
Does keyword research still matter if AI Overviews answer most questions directly?
Yes, but you need to research narrower, more specific phrasings, because broad factual questions get summarised by AI while specific ones with context or opinion attached still drive clicks to your site.
Should I stop targeting high-volume keywords entirely?
Not entirely, but treat high search volume with suspicion on purely factual questions, since those are exactly the queries AI Overviews answer directly, leaving you the impressions but not the click.
How do I know if AI search is already eating my traffic?
Check Search Console for pages where impressions have stayed flat or risen but clicks have dropped over the last three to six months, that gap is the clearest sign an AI summary is answering your query before people reach you.
What keyword types still reliably bring in clicks in 2026?
Local and transactional searches, cost and pricing questions, career and job listings, and anything needing personal judgement or comparison tend to keep pulling clicks, because AI summaries can’t replace the decision-making a human still wants to do themselves.
