Keyword difficulty: three signals that say walk away, and one that says try anyway

Every keyword difficulty score you have ever seen is a vendor's prediction, not a measurement from Google. That does not make the scores useless, but it does mean the decision to abandon a search term should rest on things you can check yourself. Three signals say walk away: the result page is owned by tools, the title-match ratio is unfavourable, and the intent is not the one you can serve. One signal says try anyway.

Read this first

This chapter assumes you already have a shortlist of candidate search terms and are deciding which ones to spend a week on. If you are still generating the list, start with how to find keywords for a new website and come back here to filter it.

It also assumes you accept the premise of the last section, which is that walking away is a legitimate outcome. Roughly two in five terms worth writing down are worth abandoning after ten minutes of checking. That figure is our own and the method behind it is below.

Where keyword difficulty scores come from

They come from the tool that shows them, computed from that tool's own link index and its own model. Google's documentation on third-party tools is explicit about the limit: "Third-party tools don't have access to our internal ranking data. They can't guarantee performance. Any predictions are their own and like predictions generally, may not happen."

The same page adds the part that gets skipped: "Google doesn't evaluate third-party services, so be wary of such claims and those making them. Keep in mind that using a service or tool doesn't guarantee ranking success." Both sentences read 2026-08-19 on the third-party SEO tools guidance at developers.google.com.

None of that is an argument for ignoring the scores. It is an argument for not treating a 42 and a 47 as different facts. Use the score as a rough band, then decide with the three signals below, all of which you can observe directly on a result page.

A difficulty score is somebody's forecast. The result page is the weather outside.

Signal one: the result page is owned by tools

Open the term and read the first two pages of results. If eight of the ten first-page entries are software vendors selling the thing the term names, the search intent has been resolved as transactional by the market, and a written answer is not what gets clicked.

This is the signal that catches the most expensive mistakes, because these terms usually look excellent on volume. High volume plus vendor-owned results is the exact shape of a term that will absorb a week and return nothing.

What you are looking for is a gap in kind, not in quality. An independent site, a forum thread, a documentation page, a personal blog — any one of those in the top ten means the page type you can produce is still eligible. Zero of them means it is not.

Signal two: the title-match ratio

Count how many pages have deliberately targeted the term in their title, and compare it to how many people search it. The mechanics take two minutes.

1. Search allintitle:"your exact term" and note the result count. This returns only pages with every word of the term in the title element, which is a decent proxy for deliberate targeting.
2. Get a monthly volume estimate from any source you have, including a free one. Precision does not matter here; an order of magnitude does.
3. Divide the first number by the second. Under about 0.25 means most of the demand is not being targeted head-on. Well above 1 means more pages are aiming at the term than there are monthly searches for it.
4. Do this before you write, not after. The ratio changes slowly, so one check lasts months.

Two cautions on the ratio, and both matter. Google's reported result counts are estimates and they wobble between runs, so treat a value near your threshold as unresolved rather than as a pass. And volume estimates from different sources disagree by multiples, which means the ratio is a band, not a number. We use it to separate obviously-crowded from obviously-open, and for nothing finer than that.

Signal three: nobody is searching it at all

The failure mode nobody plans for is not competition. It is writing an excellent page for a phrase that no human types. Autocomplete is the cheapest test available: if a term produces no suggestions in Google's own autocomplete, in either of two runs, treat that as evidence there is no standing demand behind it.

We ran that check on this handbook's own candidate list on 2026-08-19: 51 English candidate terms and 61 Chinese ones, of which 20 English (39%) and 21 Chinese (34%) returned zero suggestions. Terms as reasonable-sounding as when to give up on a keyword and seo risks of updating old content came back empty.

One caveat we learned the hard way and now repeat every time: a single empty run is weak evidence. On 2026-08-17 the same term returned nothing on the first attempt and results on the second. Run it twice before you conclude anything, and never conclude a term is dead from one blank.

The deliverable: the walk-away decision table

Read down the table. The first row that matches is the answer, and the fourth row exists because "not yet" is a real result.

The signal that says try anyway

Demand is high, the results are crowded, and one of the top ten is visibly a worse answer than the one you can write. That combination is worth a week even when the difficulty score is ugly, because the score is modelling the average competitor and you are looking at a specific one.

Be honest about what "worse" means. Older is not worse. Shorter is not worse. Worse means it does not answer the thing the searcher asked, and you can tell by reading the first screen of it and asking whether you would need to keep looking.

The term this chapter targets is itself a case in point. Autocomplete is strong for it, and the results are packed with tool vendors — signal one says walk away. We wrote it anyway, because none of those results is trying to answer "should I abandon this term", and that is the page that is missing. If you make that call, say so in your own notes, so the next person does not re-litigate it.

Where this chapter stops

These three signals rank terms against each other; they do not predict traffic. Nothing here tells you how many visits a term is worth, and we are not going to invent a conversion from position to sessions, because it varies by result page layout more than by anything you control.

The title-match ratio also degrades on very broad terms. For a two-word head term, the count returned is large and unstable enough that the ratio stops discriminating. Use it on phrases of four words or more, where it behaves.

Two more boundaries. Autocomplete reflects one country and one moment — ours runs through an exit in Japan, which shapes what comes back, and we do not know how much. And a term you walk away from today can open up later when a competitor deletes a page or a result layout changes, so a walk-away is a decision with a date on it, not a permanent verdict.

How to keep the decision

Write one line per rejected term: the term, the date, which signal killed it, and the number you saw. It takes ten seconds and it stops the same term coming back around every quarter.

Keep it in the same place as the record of what each published page targets, because the two lists answer the same question from opposite ends. That record is also what stops you writing a second page for a term you already own, which is the subject of how to fix keyword cannibalization.

Doing this check continuously rather than once a quarter, so a term that opens up gets noticed the week it opens, is what QueryWin is being built to do.

Common questions

Is keyword difficulty a Google metric?

No. Google publishes no difficulty score and states that third-party tools have no access to its internal ranking data. Every score you see is that vendor's own model.

What keyword difficulty score is too high?

There is no number that transfers between tools, because each one computes it differently. Use the score to sort your list, then decide with the result page.

How reliable is the allintitle count?

It is an estimate and it moves between runs. Treat it as an order of magnitude, and treat any value close to your cutoff as undecided rather than as a pass.

Should I chase a term with high volume and no independent sites ranking?

Not with an article. That result page is telling you which page type wins, and it is not a written answer. Build the thing it is asking for, or pick another term.

How long before I know a term was a mistake?

Long enough that pre-checking is cheaper than finding out. We do not put a number on it, because it depends on how often your site gets crawled and how much you published alongside it.