No content score we have tested predicts where a page ranks on Google.

We froze the keywords before touching any tool. For each one we pulled the live Google (us) organic top 10, kept the pages we could read, and treated their real positions as the answer key. Then we sent those same URLs through each tool’s audit with the keyword attached and recorded whatever score came back.

Within each keyword we correlated score against position, using Spearman. Read it this way: +1 means higher-scoring pages ranked higher every time, 0 means the score carries no rank information, and a negative value means it points the wrong way. We publish every keyword rather than the average alone, because an average near zero hides two different failures: steady noise, and violent disagreement.

The fixture, the ground truth and the coverage rules are all on the methodology page.

An average near zero has two possible causes. Either the score is slightly wrong in a consistent direction, which you could still work with, or it points a different way on every keyword, which you cannot. Every tool here is the second kind. Sort the keywords by correlation and every chart below crosses the axis: the same tool that tracked rank on one keyword ran against it on the next.

4 products built their scoring independently and landed in the same place. That points at the method rather than the implementation. A score computed from the text of a page cannot see links, site authority, click behaviour or intent match, and those decide most of the ordering inside a top 10.

None of this makes the tools useless. What they pull straight off a live SERP holds up: the competitor set, and the terms those pages share. The score is the part that did not survive.

The table at the top matches a tool’s competitor URLs to our ground truth on the exact path. Our ground truth came from Ahrefs, and some of its URLs came back cut short: bankrate.com/personal-finance where the page that ranks is bankrate.com/personal-finance/cost-of-living-calculator. A strict match misses those rows. A lenient rule, which counts a truth URL as matched when it is the unique prefix of exactly one URL the tool captured, rescues them, at the risk of pulling in a same-domain page that is not the one that ranked. Neither rule is the corrected value. We leave the ground truth alone and publish the correlation under both, for every run that has been recomputed that way, beside the figure each tool gets against the rank it crawled itself.

Scalenut’s sign flips between the two rules: −0.071 strict, +0.039 lenient, a move of 0.110 across zero. That is not a flaw to explain away. A correlation that changes direction when you change how a URL is matched is a correlation with no direction, which is why this page asks how far each figure sits from zero and not which way it points.

NeuronWriter’s figure in the main table stays as published, −0.052 on the strict match. Under the lenient rule it reads −0.013 over 122 pages; same sign, still inside the band.

Surfer SEO’s figure in the main table stays as published, −0.102 on the strict match. Under the lenient rule it reads −0.044 over 107 pages; same sign, still inside the band.

The last column is each tool grading its own homework, and it sits in the same place. Across every framing on this page — strict, lenient, each tool’s own rank, a pilot on its own table, and one tool's separate domain score below — no figure gets further from zero than +0.185.

We would rather be corrected than quoted. Any of the following would change what this page says.