Scalenut Review (2026): 8 Keywords, Two Answer Keys — Its Content Score Sat on Zero Against Both

The app lives at app.scalenut.com. The unit of work is a Content Optimizer analysis: one keyword, one country, one report. Each report returns a competitor table with rank, word count and content score per URL, a key-term list with an importance score out of 10 and a suggested usage range, and an editor that scores your draft as you write. The trial allows 8 of those analyses and we used 8.

We ran on the free trial, which ends 2026-08-28. The API needs a bearer token we do not handle, so the whole test went through the UI. Pricing was not recorded this round; see what we couldn’t test.

The score is the number the product asks you to work toward, so we tested it against the thing you want. For each keyword we took the real Google (us) top 10 out of our ground truth, found those URLs in Scalenut’s competitor table, and correlated the content score it gave each page with that page’s real position. Positive means higher-scoring pages ranked higher. Near zero means the score carries no rank information.

We report that correlation twice. 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 path match misses those rows. A lenient rule, which counts a truth URL as matched when it is the unique prefix of exactly one URL Scalenut captured, rescues 8 of them on this run. Strict misses real truncations; lenient can pull in a same-domain page that is not the one that ranked. So we publish both and leave the ground truth as it is.

There is a third, easier framing, and we report it too: the same score against the rank Scalenut crawled itself, over every competitor it listed. That sample is about four times larger, but the tool is marking its own work and its ranks are not Google’s.

ai for coding produced no correlation at all. Our ground truth holds only 6 web results for it, the rest of that top 10 being AI Mode and other non-page slots, and not one of those URLs matched a row in Scalenut’s table at the same path. The keyword contributes nothing to the strict mean. Under the lenient match it gains a row, which is why that framing counts 8 keywords to the strict 7.

5 of the 8 keywords rest on four matched pages or fewer, and the report marks them so nobody quotes them alone. Drop every flagged keyword and the finding survives: 3 keywords, 19 pages, mean ρ −0.100. Whichever subset you take, the answer sits on zero.

The middle row is not the true value and the top row is not the true value. They are the two ends of how far the truncation in our ground truth can move the answer, and the answer moves by 0.110 and crosses zero. On the findings page we ask how far each score sits from zero, not which way it points, and this is why.

The bottom row is the generous reading and it comes to nothing either. −0.100 over 121 rows, with 8–18 competitors per keyword, means the score barely tracks even the ordering the tool produced itself.

URLs are matched on the exact path, with the protocol, www, trailing slash and query string stripped. Same domain at a different path does not count. Of the 65 ground-truth web URLs for these keywords, 42/65 appeared in Scalenut’s tables at the same path. Not every row in those tables carries a score: 26 of its rows had none, and 10 of the matched ones were among them, which leaves 32 rows to correlate. High domain overlap, 87.2% of its top-10 domains matching Google’s, hides a lower URL overlap. The two are not the same claim.

The tables needed no cleaning: 0 duplicate rows, 0 rows without a rank. What the run did flag is in our ground truth, not in Scalenut: the truncated URLs described above. That affects every tool scored against the same truth, and the lenient column exists to show how much.

Scalenut publishes a real key-term list, so this is the same measurement we ran on NeuronWriter and Clearscope. We took its top 30 terms by its own importance score and matched them, on unigrams and bigrams, against the top 30 terms our ground truth extracted from the body text of the pages that rank. Ties on importance keep Scalenut’s own order.

About one term in six is a term the ranking pages share, and the list covers under a third of those shared terms. Across 8 keywords it found 70 shared terms in total.

The spread across keywords is wide. Best case, gpu comparison, recalled 0.400 of the shared terms. Worst case, best ai chatbot, recalled 0.133. Same tool, same settings, same day.

Do not read these against the NeuronWriter or Clearscope figures as a head-to-head. Same definition, but this run covers 8 of the 20 fixture keywords, and the other runs covered different subsets.

You can rebuild the ground truth yourself. Our Top 10 Term Finder runs the same extraction on any keyword you type: fetch the live top 10, pull the body text, keep the terms those pages share. Compare that against whatever your tool hands you.

Every competitor row carries a word count beside its rank, so we asked the same question we ask of our own corpus: do longer pages rank higher? Inside each keyword, Spearman between Scalenut’s own rank and its own word count, over 126 rows across 8 keywords: mean ρ −0.123, median −0.107, 3 keywords positive and 5 negative.

Both columns are Scalenut’s, and its word count is undocumented, so this row sits beside our corpus and the other tools’ tables on the length-versus-rank finding and is never pooled with them. Four sources, four numbers, no average.

Every tool runs against the same frozen fixture and the same ground truth (serp-truth-v1 + terms-truth-v1): the live Google (us) top 10 for each keyword, the body text of those pages, and the terms those pages share. Country, language and device stay constant across tools. The full procedure is on our methodology page.

Scalenut’s API needs a bearer token, which we do not handle, so we drove the app through its UI and read the competitor table and the term list out of the rendered report’s component state rather than off the page’s markup, whose class names change with every build. The keyword label on each capture came from the report’s breadcrumb, because the component state does not expose it; we checked all 8 against the fixture character by character before computing anything. One extractor version, 8 captures on v1. The captures recorded the country setting as US.

Before publishing, an independent script recomputed every headline number from the raw captures without importing the run’s code, and wrote the strict and lenient score-to-rank figures to a sidecar file that this page reads. The scripts are scripts/tests/content-opt/scalenut/{extract_report.js,run.py}. Last tested 2026-08-21, on the free trial, against app.scalenut.com.