Someone told me recently that keyword research is dead. They said it with the confidence of a person who has not had to explain a traffic drop to a client.
Keyword research is not dead. What has died is the assumption that the phrases in your keyword tool are the phrases deciding whether you get seen.
Because when a person asks an AI a long, messy, conversational question, the system does not look for that phrase. It breaks the question into parts and looks for the best answer to each part.
Those parts are the new long tail. Almost none of them appear in a keyword tool, because nobody types them and no volume is recorded against them.
Here is how I find them anyway.
Quick Info
Why the Long Tail Got Longer
Two things happened at once.
Voice and conversational interfaces made people phrase queries as full questions rather than keyword fragments. Then AI search made those long questions actually work, because the system can decompose them instead of matching them literally.
The result is a shift in where the intent lives. A search for best crm is a category query. A question about which CRM suits a two-person consultancy that already uses Google Workspace and hates data entry is five sub-questions wearing a coat.
The long tail is no longer a rarer phrase. It is a more specific situation.
This is genuinely good news for small sites, because specificity is the one thing you can beat a large competitor on. They write for the category. You can write for the situation.
Method One: Read the Answer, Not the Tool
This is the technique that has replaced half my keyword research, and it costs nothing.
Take the big question in your niche. Ask it in AI Mode, and in one other assistant. Then read the answer structurally rather than for content.
What sub-topics did it choose to cover? In what order? What comparisons did it make? What caveats did it add? Which sources did it credit for each part?
That structure is a brief written by the system you are trying to appear in. Every sub-topic it covered is a heading you should own, and every source it credited is a page you should read to see why it got picked.
I do this for twenty questions a quarter and it produces a better content plan than any tool I subscribe to.
Method Two: Mine Your Own Search Console
Search Console is still the only source of queries people genuinely used to find you, and most people never filter it properly.
Then group what you find by situation rather than by phrase. Ten queries describing the same person in the same predicament are one page, not ten.
Method Three: Sources Nobody Automates
The best long-tail questions come from people, not tools, and there is no shortcut here.
I keep a running note on my phone for these. It is the least sophisticated part of my process and consistently the most useful.
How to Actually Use Them
This is where most people go wrong, because the old approach was one keyword, one page, and that no longer scales when the tail is this long.
Do not build a thin page per question. Build a substantial page per situation, with a self-contained section per sub-question.
One page built this way can appear in dozens of AI answers, because it is not competing for one query. It is a library of quotable passages on one coherent topic.
What to Stop Doing
Three habits that made sense five years ago and now waste money.
Stop filtering out zero-volume keywords. Zero volume in a tool increasingly means the phrase is a sub-question rather than a typed query, and those are precisely the ones worth owning.
Stop building separate pages for near-identical phrasings. You are competing with yourself and diluting a topic that should be consolidated.
Stop optimising for exact-match phrasing inside the copy. Research into generative search consistently finds keyword repetition adds little and can hurt, while statistics, quotations and named sources help substantially. Write for the meaning, not the string.
Frequently Asked Questions
Yes, but as validation rather than discovery. Tools confirm demand and competition. Discovery increasingly comes from reading AI answers, your own Search Console and real customer questions.
Often yes. Many are legitimate sub-questions that AI systems ask on the user's behalf, and they face very little competition.
Long enough to cover the situation properly, structured into self-contained sections of roughly a hundred and fifty words. Total length follows from the number of real sub-questions, not from a word target.
Generally yes, because specificity correlates with intent. Someone describing their exact situation is closer to a decision than someone searching a category.
The process where an AI search system breaks one question into several underlying searches, gathers results for each, and synthesises an answer. It is why sub-questions matter more than head terms.
That is now the goal. A well-structured page covering a situation thoroughly can be cited for dozens of related questions.
Before You Go
The shift here is small in method and large in results. Stop looking for phrases with volume, start looking for situations with specificity.
My question-logging template and the research process I use each quarter are in the resources section, and there is a full module on it in my SEO and AI search course.
If you want a content plan built this way for your own business, that is what I do.
Situations, not strings, Tariq
I'm a marketing consultant, entrepreneur and content creator. I help businesses grow through practical marketing, websites, SEO, content and AI.
More About Tariq →