The ChoiceScore Index — August 2026

Original data study · ChoiceScore Research Desk · first edition, updated quarterly.

Since launch, ChoiceScore has generated and analysed 3,407 public decision reports. In August 2026 we rebuilt the library around a strict human-curation rule: only 38 reports — about 1.1% — earned a place in search. This index publishes what that exercise taught us. The dataset is our own production database; these numbers exist nowhere else.

Finding 1: machine quality scores could not pick the library

Every report receives an automated 8-dimension quality score (0–100). The counter-intuitive result: the pool our editors rejected averages a higher machine score (90.7) than the pool they approved (87.0). Automated scoring rewards completeness and structure; it cannot see whether a page genuinely helps a person facing the decision. This is why automated scores now only prioritise review at ChoiceScore — they never approve indexing.

Finding 2: “it depends” is the honest verdict half the time

Across the 38 curated reports, 19 verdicts are depends, 9 recommended, 6 uncertain, and 4 high-risk. Exactly half of high-consideration decisions resolve to “it depends on inputs you control” — which is why every report exposes its assumptions instead of forcing a yes or no.

Finding 3: make-money ideas score worst

The lowest Choice Scores in the library cluster in one place: dropshipping (47), new affiliate blogs (49), YouTube channel economics (50), Airbnb short-term rentals (51), and coffee shops (54). By category, small business averages 51.8 and website success 50.3 — far below head-to-head comparisons (73.8) or cost-of-living moves (69.5). The evidence keeps saying the same thing: buying decisions are easier to get right than income-generating ones.

The data

Score distribution: 2 curated reports score in the 40s, 8 in the 50s, 15 in the 60s, and 13 in the 70s. No curated report scores above 79. If a decision tool hands you a confident 95/100, our data suggests you should ask what it is not counting.

Method and limitations

Counts and averages were computed directly from our production database on 2 August 2026 (curated n=38; rejected public pool n=3,369). Choice Scores are our own 0–100 model outputs, not market data. Category averages with n≤4 are indicative only. We re-run this index quarterly and note material movements; errors can be reported via our corrections policy.

Cite as: “ChoiceScore Index, August 2026 — getchoicescore.com/choicescore-index”.