Every keyword tool shows a difficulty score, usually 0–100, usually colour-coded so that green means go. It is the most quoted number in SEO and the most widely misread.
The first thing worth saying plainly: keyword difficulty is not a Google metric. Google does not publish one, does not endorse one, and does not use any vendor's score in ranking. Every difficulty figure you have ever seen is a vendor's model, built from what that vendor can observe — mostly the backlink profiles of the pages currently ranking. Two tools will give you two different numbers for the same query, and neither is wrong, because they are answering slightly different questions.
What the number is actually modelling
Most difficulty scores are, at heart, a summary of one thing: how much link authority do the current top-ranking pages have? A query where the first page is ten established domains with thousands of referring domains each will score high. A query where the first page includes a forum thread, a two-year-old blog post and a thin category page will score low.
That is genuinely useful information. It is a reasonable proxy for "could a page from a site like mine plausibly appear here at all?" — which is a real question, and one worth asking before you spend a day writing.
But notice what the model does not contain:
- Whether the query is one you can answer better than the incumbents. A low score on a topic you know nothing about is a trap, not an opportunity.
- What the searcher actually wants. Difficulty says nothing about intent.
- Whether the result is dominated by something you cannot compete with — a Google feature, a product carousel, a video block, an official source.
- Whether the query is worth anything to your business.
A difficulty score is a filter for one failure mode — being outgunned — and blind to every other one.
Why we filter below 25
UNC AI SEO's research step defaults to keyword difficulty below 25 and monthly volume near 1,000. Those are not magic numbers and we would rather explain them than have them taken on faith.
The difficulty ceiling exists because for a site that is still building its authority, the pages that rank for high-difficulty queries are usually not beatable by writing something better. They are held in place by years of accumulated links. You can write the best page on the internet for that query and still sit on page four. Filtering low is an admission about where the site currently stands, not a claim about content quality.
The volume figure is the other half. Very high volume tends to correlate with commercial competition and with queries broad enough that the intent is unclear. Very low volume — a handful of searches a month — means that even a first-place ranking changes nothing measurable. Something in the region of a thousand searches a month is often specific enough to have a clear intent behind it and large enough to notice when it works.
Both numbers should move as a site's position changes. A domain that has been publishing consistently for two years and has earned real citations has no business still filtering at 25. Treat the defaults as a starting position, not a permanent setting.
The three checks worth more than the score
Once a keyword clears the numeric filter, the number has done its job and stops being useful. What follows matters more.
1. Open the results and read them
Not skim — read. You are looking for what the ranking pages have in common, because that consensus is Google's own read on the intent behind the query. If every result is a step-by-step tutorial and you were planning an opinion piece, the query is not asking for your opinion.
You are also looking for the gap. What does every one of these pages fail to say? What question would a reader still have after reading all of them? That gap is the only real reason to publish. If you cannot name it in a sentence, you do not have a page yet.
2. Check what you can honestly claim
Google's guidance on helpful content asks, among other self-assessment questions, whether the content demonstrates first-hand expertise and depth of knowledge — whether it is written by someone with demonstrable experience of the subject. This is not a box to tick. It is a question about whether you should be the one writing.
If the honest answer is that you would be paraphrasing the same four sources everyone else paraphrased, the difficulty score is irrelevant. You would be adding another page that says what is already said. Pick a different topic, or go and acquire the experience first.
3. Check the query is worth winning
A keyword can be easy, well-intentioned and completely disconnected from anything your business does. Traffic that never becomes a reader, a subscriber or a customer costs you the same effort as traffic that does.
This is the check most often skipped, because tools do not measure it and it requires knowing your own business. Ask what a person arriving on this page is likely to want next, and whether you have anything to offer them.
Missing data is better than guessed data
One deliberate behaviour in our own research step is worth explaining: when a provider returns no difficulty or no volume for a keyword, we exclude it rather than substitute an estimate.
It would be easy to fill the gap — infer a figure from similar queries, or show a blank as a zero. Both would produce a tidier table. Both would also mean a decision made on a number that nothing supports. A visibly missing metric prompts a human to go and look; a plausible fabricated one does not.
What to take away
Difficulty scores are a coarse, useful, widely misunderstood filter. Use one to remove queries you cannot realistically compete for. Then close the tool, read the results, and decide whether you have something worth adding.
The number tells you whether you could rank. Only you can answer whether you should publish.