Content optimization

Content optimization tools, compared by test

One frozen keyword fixture, one ground truth, the same five measurements. Rows fill in as each tool finishes testing.

How each column is measured, and what it does not tell you, is set out in our methodology.

ToolStatusTerm precision@30Score ↔ rank (ρ)Cheapest paid plan
Frase tested2026-08-180.12*-0.12$39/mo
NeuronWriter† tested2026-08-210.23-0.05$23/mo
MarketMuse queued———
Scalenut§ tested2026-08-210.16-0.07§—
Surfer SEO tested2026-08-240.16-0.10—
Clearscope‡ tested2026-08-200.200.11‡$129/mo

← swipe the table sideways for the rest of the columns

* On Frase’s precision@30

Frase publishes no term list, so its precision@30 is a substitute measure: its keyword research output scored against our term ground truth. A low value there means the tool does not do that job, not that it does it badly. How we measured it.

† On NeuronWriter’s two rounds

The NeuronWriter run covers all 20 fixture keywords, the same set as the Frase run, captured in two rounds (15 on 2026-08-19, 5 on 2026-08-21) and recomputed as one run under one set of definitions. The two precision figures now rest on the same keywords, though Frase’s is still a substitute measure (see *). The ρ figures are not a pair: Frase’s comes from 5 keywords and 38 pages, NeuronWriter’s from 18 keywords and 110 pages. The date in the row is the day the second round completed; the first round was 2026-08-19. How we measured it.

‡ On Clearscope’s pilot

Clearscope’s free trial allows 3 reports and each keyword costs one, so that row is a pilot: 3 keywords, tested 2026-08-20. Its ρ is also a different measurement from the Frase and NeuronWriter figures. Clearscope grades the search results itself, and we correlated those grades with the positions in its own competitor table rather than with the frozen Google (us) truth the other rows use. That is the easier version of the test. Read the row on its own; it does not belong beside the others. How we measured it.

§ On Scalenut’s sample and its two answer keys

Scalenut’s trial allows 8 analyses, so that row covers 8 keywords, tested 2026-08-21, with 8–18 scored competitor rows per keyword against 25–30 on the NeuronWriter run. Same framing as NeuronWriter’s ρ, smaller sample; do not rank the two. The figure in the cell, −0.071, is the strict URL match against our frozen Google (us) truth, the same rule the NeuronWriter cell uses. Our truth holds some URLs Ahrefs returned truncated; rescue those and Scalenut’s ρ is +0.039 over 40 pages. The sign flips, the size does not, and the review shows both. How we measured it.

¶ On Surfer’s ordering, second score and day-two re-run

Surfer’s Pro trial covered the whole fixture: 20 keywords, tested 2026-08-24. The ρ in the cell, −0.102, is its 0–100 content score against our frozen Google (us) truth on the strict URL match, the same rule the NeuronWriter cell uses; rescue the truth URLs Ahrefs returned truncated and it reads −0.044 over 107 pages — same sign. Its table also carries a 0–10 domain score the cell does not read (+0.185 against the same truth). Its precision@30 scores the first 30 terms of its panel in Surfer’s own display order, because the tool publishes no numeric term importance, where the other term lists are ranked by their own importance values. And a re-run of 5 keywords 25.61–25.81 hours later returned term lists below our 0.7 stability threshold on 3 of 5, while the scores held — so read the precision figure as dated 2026-08-24. How we measured it.

Head-to-head

Two tools at a time, on whatever both runs measured. Each page carries the sample size and the run date behind every number.

A category on one page

Same rows, arranged by the question you are asking — the term list, the competitor set, the price, the claims in a generated draft — with what each run did on each one.

Reading the columns

ρ is the mean Spearman correlation between a tool’s content score and actual Google position. Positive means higher-scoring pages rank higher; near zero means the score carries no rank information. Term precision@30 is how many of a tool’s top 30 recommended terms appear among the terms the current top-10 pages actually share. A dash means not yet tested. We do not publish estimates.

Every score we have tested landed on zero

We scored Frase, NeuronWriter, Scalenut and Surfer against the real Google (us) top 10, and no score told us anything about where a page sat in it: -0.115, −0.052, −0.071 and −0.102, the Scalenut figure becoming +0.039 and Surfer’s −0.044 once truncated truth URLs are rescued. Clearscope’s pilot took the easier test, its own grades against its own ranking, and came to +0.108 on 3 keywords; Scalenut and Surfer under that same easier framing came to −0.100 and −0.084. Five products, five scoring models built independently, nothing to separate any of them from zero. We do not average across those measurements and neither should you. But if you are choosing between products on the strength of their score, the evidence so far says the score is not the thing to choose on.

Why so many dashes

Each row costs a full test run: a frozen keyword fixture, a ground truth rebuilt from the live SERP the same day, and every raw response kept on our side. We publish a row when the numbers exist, not when the tool launches a campaign. If you want a specific tool moved up the queue, say so: hello@toolverdict.ai.