Are Keywords Still Relevant in SEO? What Changed, and What Didn’t

Yes. Keywords still decide which page Google shows for which search.

What changed is that they stopped being strings to match and became questions to answer. That sounds like a distinction without a difference until you look at what people actually search, which this guide does using real data from this site.

“Keywords are dead” is shorthand for something narrower and more useful than it sounds. Worth unpacking, because the people saying it and the people arguing with them are usually talking about different things.

The short version

• Keywords matter. Keyword matching does not.
• Google treats different wordings of one question as one question. Your job is the question, not the wording.
• Density, exact match and keyword counts are the parts that died.
• Keyword research is more useful than ever, as a map of what people want, not a list to insert.

What “Keywords Are Dead” Actually Means

Nobody serious means that search terms stopped existing. People still type things into a box, and Google still decides what to show them.

What died is a specific set of practices from when search engines matched text literally:

• Keyword density targets. Hitting 1% or 2% of a document.
• Exact-match repetition. Using the precise phrase because a variation would not count.
• A page per variation. Separate pages for “cheap X”, “affordable X” and “budget X”.
• The meta keywords tag. Ignored for well over a decade.
• Keyword stuffing. Now a documented spam policy rather than a tactic.

All of those worked, briefly, because the machine was simpler than the language. The machine caught up.

What Replaced Matching

Google resolves meaning rather than text. It identifies the things being discussed, the relationships between them, and the question the searcher is actually asking.

Three practical consequences follow, and they are the whole of modern keyword work.

1. Wordings collapse into questions

“How much does SEO cost”, “SEO pricing” and “what do SEO agencies charge” are one question in three costumes. Google knows that. Building three pages splits your own signals and creates competition between your URLs.

2. Pages rank for things you never targeted

Because Google understands the page rather than indexing its phrases, a page that answers a question well will surface for phrasings you never wrote down.

3. Intent separates pages, not vocabulary

“Best CRM” and “buy CRM software” use different words and, more importantly, want different things. One reader is comparing, the other is purchasing. That is two pages.

Meanwhile “best CRM” and “top CRM tools” want the same thing. That is one page. Vocabulary is not the dividing line; intent is.

Then and now, side by side

Every row on the left was a real technique that worked. That is why the advice persists: it is not made up, it is expired.

What This Looks Like in Real Data

Here is one page on this site, over three months in Search Console. It is an article about programmatic SEO on Webflow, and it was written to answer one question.

Look at the first two rows. They are the same three words in a different order. Google records them as separate queries and serves the same page for both, with almost identical volume.

Nobody optimised for both. There is no version of keyword work where you write one page for “programmatic seo webflow” and another for “webflow programmatic seo”, and under the old rules, that is exactly what people did.

The fourth row is the same point stretched further: a long, specific query nobody would have thought to target, matched to the page because the page answers it.

The honest caveat

Four queries is a small number. You will see claims that one page ranks for hundreds of keywords, and on an established site that is true.

This site averages position 56.9 and gets around 1,100 impressions a quarter, so its pages surface for a handful of phrasings rather than hundreds. The pattern is identical; the scale follows authority.

That is worth saying because the pattern is the point, and it holds at any size. A page answering one question will collect the variations of that question, however many there turn out to be.

Where Keywords Still Matter Exactly

Four places, and the list is short on purpose.

• Deciding what to write. The most valuable use. Keyword data tells you which questions people actually ask and roughly how many ask them, which stops you writing for an audience of nobody.
• Choosing the phrasing. If people search “site not showing on Google” and your page says “de-indexation issues”, you have written accurate content for a query nobody enters.
• Placement, once. Title, H1, first paragraph, and naturally through the body. There is more in where to add keywords, including the placements that do nothing.
• Spotting overlap. Two pages after one question is a real problem, and keyword data is how you find it.

What is not on that list: counting them. How many to use per page is a question with a simpler answer than people expect.

Keywords and AI Search

This is where the “keywords are dead” claim has picked up new energy, and it deserves a straight answer rather than either extreme.

What actually changed

People ask assistants in full sentences, conversationally, and often across several turns. Nobody types “best crm small business” at ChatGPT; they describe their situation.

So the input looks less like a keyword than ever. But the assistant is still resolving a question, and it still has to decide which sources answer it.

What did not change

• Questions still have topics. A conversational question is still about something, and that something is what your page needs to cover.
• Retrieval still needs a match. When an assistant searches the web mid-answer, it issues queries. Those queries look a lot like keywords.
• Clear pages get quoted. A page that resolves one question cleanly is easier to lift from than one covering six.

The shift worth making is from phrases to questions, which is the same shift Google’s own move to meaning already required. Being present in AI Overviews follows from the same work, not a separate technique.

What is genuinely new is measurement. Rankings and clicks now diverge, because a page can hold its position and lose the click to an answer above it.

How to Do Keyword Work Now

The process changed shape rather than disappearing. Five steps, and only one of them involves a tool.

• Start with a question someone actually asked you. A customer, a support email, a comment. Real questions beat invented ones because someone already needed the answer.
• Find how it is phrased in the wild. Autocomplete and People Also Ask give you real wording for free. Note the phrasings; do not plan a page for each.
• Check the results page before writing. What ranks tells you what Google thinks the question means, and what format it expects. This is the step people skip and then wonder why a good page did not rank.
• Check you do not already cover it. If a page already answers it, improve that page. Adding a second one creates competition with yourself.
• Write the answer, then place the phrasing. In the title, the H1, the opening, and wherever it belongs naturally. The on-page checklist covers the placement in order.

Notice that volume data appears nowhere in the first four steps. It is useful for prioritising between questions, not for deciding whether a question is real.

Five mistakes that follow from old advice

• A page per phrasing. The most expensive one, because it creates competing pages that then need merging.
• Writing to a density. Produces text with the phrase wedged into places it does not fit.
• Choosing by volume alone. A high-volume query owned by Semrush and Moz is not an opportunity, whatever the difficulty score claims.
• Ignoring the format of the results. Publishing a list where the results are all tutorials, or a service page where they are all comparisons.
• Treating tool difficulty scores as fact. A keyword export used on this site rated “seo services” at 45,700 monthly searches with a difficulty of 1. Open the results page and look.

When Keywords Matter Less

Being honest about the limits is more useful than defending keywords everywhere.

• When the page is not indexed. No keyword work reaches a page Google has not kept. Check whether it was crawled and declined first.
• When the site has no authority. A perfectly targeted page on a site nobody links to still sits on page five. That is the binding constraint on many small sites, including this one.
• For brand and navigational searches. People searching your name will find you regardless.
• When the format is wrong. If the results are all comparisons and you published a service page, no amount of phrasing fixes it.

Should You Still Track Keyword Rankings?

Yes, with two adjustments.

Rankings are still the clearest signal that a page is gaining or losing ground, and they move before traffic does. Dropping them entirely means finding out later.

The adjustments:

• Track questions, not phrase lists. Watching twelve variations of one question produces twelve lines that move together and tell you one thing.
• Pair position with impressions and clicks. A position that holds while clicks fall is the AI-answer pattern, and a rank tracker alone will not show it.

Search Console is free and shows all three together. Dedicated rank tracking tools add daily granularity and competitor tracking, which matters more the more competitive your space is.

The highest-value use of ranking data is finding pages between positions 11 and 20, striking distance, where Google has already accepted relevance and a small improvement can move the page.

Frequently Asked Questions

Are keywords still relevant in SEO?

Yes. Keywords still determine which page Google shows for which search. What stopped working is keyword matching, density targets, exact-match repetition and a separate page per phrasing.

Does keyword density still matter?

No. There has never been an official target and there is nothing to measure against. The only real failure is using a phrase where a person would not have, which reads as stuffing.

Do I still need keyword research?

More than before, but for a different purpose. It tells you which questions people ask and in what words, which is how you decide what to write. It is not a list of phrases to insert.

Do keywords matter for AI search?

The questions matter; the exact phrasing matters less. Assistants resolve questions and still issue web queries when they search mid-answer. A page that answers one question cleanly is what gets cited.

Should I make separate pages for keyword variations?

No. Google treats different wordings of one question as one question. On this site, “programmatic seo webflow” and “webflow programmatic seo” both returned the same page with near-identical impressions. Separate pages would only have competed with each other.

Is keyword stuffing still a penalty?

It is a documented spam policy, so yes. In practice most sites do not stuff deliberately; they end up with awkward repetition by following density advice.

Should I still track keyword rankings?

Yes, but track questions rather than long lists of variations, and read position alongside impressions and clicks. A stable position with falling clicks usually means an AI answer now sits above you.