Best words for SEO, how to choose them and the one question a keyword tool cannot answer

There is no list of best SEO words and anybody selling you one is selling a list that is wrong for your site by construction. A word is good or bad only in relation to who you are and who already holds it.

The four part test that replaces the list

Run every candidate word through these four in this order. Most candidates die at step two, which is the point.

One, does the word describe a problem or a product. Problem words are searched by people who do not yet know what to buy. Product words are searched by people who do. Both are useful and they need completely different pages. The most common waste in this whole discipline is a product page aimed at a problem word, which ranks for nothing and converts nobody.

Two, who holds the first five results, and can you credibly be there. This is the step that kills most lists. Open the results page and look at who is on it. If the first five are national publishers, comparison sites and two competitors ten times your size, that word is not a plan, it is a wish. If the first five include a forum thread, a weak page from 2019 and two sites your size, that is an opportunity and it is worth more than a bigger volume figure elsewhere. Volume tells you the size of the prize. The results page tells you whether you can win it. Only one of those is on the tool's screen.

Three, can you say something the current results do not. If you cannot answer better than what is already ranking, you are hoping to win on authority you do not have. Where you can answer better is usually where you have information nobody else does, which is your own data, your own prices and your own cases.

Four, would a buyer searching it ever buy from you. The step everybody skips. Some high volume words are searched almost entirely by students, competitors and people looking for a free version. Ranking for them produces traffic, a healthier looking dashboard and no revenue.

The three word shapes worth most to a small site

The specific long phrase. Four or more words, low reported volume, a buyer who knows exactly what they want. Individually small, collectively most of the opportunity for anybody who is not already large.

The word plus the place. If you serve an area, the area belongs in the word, and this is the one category where a small business routinely outranks a national one.

The comparison. One thing against another, or alternatives to a named product. Searched late in a decision by somebody with a budget. Harder to write honestly than it looks, which is exactly why there is room.

What the tools report, and what the number actually is

Keyword tools are worth paying for and it is worth knowing which part of their output is measured and which is modelled. Semrush is unusually candid about this on its own page and we read it at semrush.com/features/competitor-analysis on 21 October 2026.

It describes its traffic data as coming "from real clickstream data, anonymized records of over 200M internet users' online journeys", "sourced from a large panel of third-party providers updated every one to two days", then "cleaned with a proprietary machine learning model that filters anomalies". It publishes its scale as "1B+ daily global clickstream events" and "808M+ domain profiles".

That is a careful and honest description of an estimate built from a very large sample. Read it that way. A volume figure is a modelled estimate, not a count, and the right use of it is to rank candidates against each other rather than to forecast anything.

The demand no keyword tool reports, and this is the part that matters now

Here is a hard fact about this category, and it applies to every keyword tool including the best ones.

When a buyer asks ChatGPT, Gemini, Perplexity, Claude or Google AI Overviews who to buy from, that question is not a search query. It does not enter a keyword database, it has no volume figure, and no planner will ever show it to you. The answer it produces is a short list of names. If your business is not among them, there is no position to improve and no word to optimise, because there was no results page.

So the keyword list, run properly as above, is still the right way to plan pages. It is simply no longer the whole picture, and the missing part is invisible to the instrument rather than small.

AI Knows Us measures that missing part and nothing else. We are not a keyword tool, we do not report volumes and we are not an alternative to Semrush, Ahrefs or Keyword Planner for choosing words. We take the buying questions your market asks, put them to the five surfaces, and report whose names come back. Our ladder at aiknowsus.com/llms.txt, read 21 October 2026, publishes Growth at $299 a month with 100 buyer questions watched, 1,600 credits and all five engines. Note the unit carefully, because it is not a keyword unit: it is a buyer question, asked on an app, once. And note the floor honestly, because the lower number does not do this job: SEO Start at $149 publishes 0 questions, 0 credits and 0 engines.

What to read next

If you want the tool rather than the method, the free long tail options and where each one's data actually comes from are compared at best free long tail keyword tools compared. We link to it deliberately. When we measured inbound links across 2,286 stored page bodies on 21 October 2026, no other page on this site links to that one, and 939 of our 2,200 live pages were in the same position, 42.7 percent, an upper bound.

If you would rather hand the whole job over, what to require as a deliverable before paying is at best keyword research services and agencies compared.