Methodology: How We Verify Pricing and Rank Tools

How we select official pricing sources

We prioritise each provider's own published pricing page. Where a provider blocks automated access, we use a trusted aggregator that mirrors official pricing and updates promptly. Every source is recorded with a trust level, and only official or trusted sources are used to update the prices shown on the site.

How pricing is checked

Prices are fetched, classified, and reviewed through our pricing intelligence pipeline. High-impact or lower-confidence changes are escalated for human review before they are published. We store the source URL and the date each price was last checked.

What Live, Recent, and Stale mean

Every pricing surface carries a freshness state based on when the price was last verified:

• Live — verified 3 days ago or less
• Recent — verified 14 days ago or less
• Stale — verified more than 14 days ago

How changes are recorded

Every reviewed pricing change is written to our pricing history and changelog with the tool, the exact fields that changed, the previous and new values, the source, and the date detected. Nothing is silently overwritten.

How comparisons and recommendations are produced

Comparisons and rankings are based on real, current pricing combined with quality and cost-efficiency scores and the specific job the tool is being bought for. Recommendations aim at best value for a stated use case — not the most expensive option.

How affiliate relationships work

We earn affiliate commissions from some outbound links, at no extra cost to you. Affiliate relationships are disclosed — see our affiliate disclosure. Rankings are never sold. Providers cannot pay to change rankings, add positive coverage, or alter the pricing we show.

What providers cannot pay to change

• Rankings and the order tools appear in
• The prices we display
• Whether a tool is recommended for a use case
• Positive or negative editorial coverage

How the AI Waste Score is calculated

The free AI Waste Score is a self-reported diagnostic, not a billing export. Eight answers map to a raw rubric (maximum 95 points). The public card shows that raw total scaled to 0–100 so a share line can say “68/100”.

• Overlap of two tools doing the same job is the heaviest leak (20 raw points).
• Unused weekly tools, 4+ paid tools, and $100+/month spend are 15 raw points each.
• Team seats, dual subscription + API spend, and low plan confidence are 10 raw points each.
• Bands stay on the raw rubric: Low ≤30, Medium ≤65, High above 65 — then displayed as 0–32 / 33–68 / 69–100.

The seven leak areas on /ai-cost-audit (unused seats, duplicate tools, plan mismatch, cheaper alternatives, renewal risk, API cost exposure, agent permissions) are the leaks we see most often in buyer reviews. A persona selector only reorders those areas. It does not change prices or rankings.

The “potential monthly waste” range is a banded fraction of the spend you typed — a planning range, not an invoice. Live catalogue timestamps on the result card come from the same models.json freshness used everywhere else on the desk.

How the content-quality engine works

The free Founder Copy Review is a review gate, not a publisher and not an AI detector. Google’s helpful-content systems look for unhelpful, stale, and scaled filler. The desk scores the same failure modes Andy actually edits for.

• Helpfulness (22): a direct answer, a segmented verdict, dates and sources — not a keyword essay.
• Anti-slop (20): brochure phrases, padding, and interchangeable launch-post language.
• Integrity (18): no invented crowds, no sold rankings, no decorative timestamps.
• Architect tone (16): functional fit before price; cost of an accepted outcome.
• Andy voice (12): measured judgment. The goal is a decision-worthy page, not beating a detector.
• Freshness (12): Live ≤3 days on money pages; Recent ≤14. Other families use 14 / 30. Overdue clocks go to the refresh queue; they do not invent a “do not ship.”
• Andy publication gate (14 checks): reader, answer early, fit before price, entry vs sustained, scale, accepted outcome, claims vs judgment, uncertainty, ranked call, alternatives, less uncertainty, no jargon, one practical insight, affiliate last.

The public-figure board (Buffett, Munger, Jobs, Bezos, Graham, Feynman, Lovelace, Godin) is a set of editorial lenses. Those lines are written by the desk. They are not quotes, endorsements, or reviews by the people named.

D1 content-autopilot uses only the hard gate — invented proof, sold rankings, freshness theatre, and dense slop. A dry source-backed patch is not rejected for missing voice phrases. Nothing in this engine writes models.json or ships a page on its own.

Known limitations

• Prices can change between our last check and your visit — always confirm on the provider's page before buying.
• Token-based API costs depend on your own usage; calculator figures are estimates.
• Some providers restrict automated access, so certain prices update less frequently.

How to report incorrect pricing

If a price looks wrong, tell us the tool, the plan, the price we show, and the official source. Use the contact page — we review every report and correct verified errors.