A keyword difficulty score on its own tells you almost nothing. The same number means different things from different vendors, and the same vendor gives a different number next month. Before you act on one, write down who scored it, on which date, and what you compared it against.
Keyword difficulty, KD for short, estimates how hard it is to get a page into the top ten Google results for a specific term. Most platforms put it on a 0 to 100 scale, higher meaning stronger competition. Google publishes no such figure. Every score you will ever see is a vendor's model of the results page, and vendors disagree on the bands, on the inputs, and sometimes on the same keyword on the same day. That is why people shop for it as a checker with a vendor's name attached.
What the Bands Mean, and Where Vendors Split
Two band maps circulate for the same scale, and the labels do not line up.
The common one reads 0 to 29 as easy, 30 to 49 as medium, 50 to 69 as hard, and 70 to 100 as very hard, with the harder bands calling for stronger content, more incoming links, and a stronger domain. Semrush cuts the scale into six: 0 to 14 very easy, 15 to 29 easy, 30 to 49 possible, 50 to 69 difficult, 70 to 84 hard, 85 to 100 very hard, and layers a separate personal difficulty for your own domain on top of the general score.
The consequence is small and real. A term in the low thirties is medium on one map and merely possible on the other, and anything near a boundary changes meaning when it crosses vendors. When you log a score, log the vendor's band map beside it, so a later reader can tell whether medium meant 30 to 49 or something else.
What Goes into the Number
The part every vendor agrees on is links and authority.
Difficulty formulas are built chiefly from the backlinks pointing at the pages that currently rank: the median number of referring domains (distinct websites linking to a page), the share of those links that pass ranking credit versus the ones marked nofollow, which do not, and the authority of the competing domains, a vendor's estimate of how much a site's links are worth. More quality referring domains behind the current top ten, and stronger domains among them, means a higher score. Vendor methodology pages and independent explainers agree on this core.
Semrush states four further inputs: search volume, with higher-volume terms usually scoring harder; features on the results page such as featured snippets, People Also Ask boxes and local packs, which raise the difficulty of winning a click; branded terms, which are harder for outsiders because the brand owns the results; and word count, with longer long-tail terms usually scoring lower. Use the links-and-authority core as the shared explanation across vendors. Treat the four extensions as one vendor's stated method, because no second vendor documents them as general mechanics.
How Agent Builders Actually Use It
Among people wiring SEO data into agents, difficulty is a filter inside a prompt, not a definition.
One prompt that circulates keeps keywords with difficulty under 40, monthly volume between 300 and 50,000, and informational or commercial intent, then ranks the survivors with the author's own opportunity score:
volume × CPC × 0.4 + (100 − KD) × 0.6
with each pick tagged Quick Win, Core Revenue or Long Game. CPC is the cost per click advertisers pay on the term, used here as a proxy for how much the traffic is worth. The weights are one builder's choice. Borrow the shape, demand against ease with a label for what each pick is for, and set your own weights.
The loop around that prompt is always the same: pull live volume and difficulty, combine them with what the company sells and what it has already published, and only then recommend a topic. Builders cache the responses per keyword and location and remove duplicates before the next batch, and some now check volume, difficulty and CPC across more than one data source instead of trusting one vendor's keyword database. The difficulty number is one operand among several, always dated, always cached. Where the keywords come from in the first place is a separate step, and suggest discovers while volume validates is the split most of those loops run on.
Why the Same Keyword Scores Differently
Expect mismatch and plan for it.
Vendor-comparison threads carry the claim that DataForSEO underestimates difficulty because it crawls less of the web, usually with the advice to trust Semrush instead. Nobody in those threads publishes a benchmark, so read the claim as the reason mismatch threads exist, not as a settled fact about either crawl. The grumbling about volume accuracy around the same vendors sits at the same level: evidence that a gap exists, never a number you can compute with.
A workable posture has three parts.
Record the vendor and the date with every score. A 38 from one vendor this month is not the same claim as a 38 from another vendor in March. qequals does not score difficulty at all; what it keeps is the dated, weekly snapshot of the live results page, which is the thing you would otherwise have to save by hand next to each score.
Check picks against your own ground truth. What did your own pages do on the queries you already rank for, in Search Console? That tells you more about which vendor's model fits your site than any argument about which model is right in general.
When you switch sources, say so in the log. Checking across vendors is a coverage choice with its own trade-offs; it does not reveal a single true score that one vendor had wrong.
One widely read thread calls keyword difficulty a nonsense metric outright. Read it as the end state of a score without provenance. A number with no vendor, no date and no baseline deserves that skepticism. A number with all three is a working input.
