Skip to the charts

How we grade what each tool covers

What each grade means, what we read, how two readers settle a grade, and when a maker's reply changes one.

Method 1.0, in use since 19 November 2026.

What a grade says

Each cell of the grid is one tool and one row. The graded rows are the AI engines, one by one and all five at once. The grade says what the tool's cheapest plan with a public price covers, as its own pricing page puts it on the day we read it. It doesn't say how well the tool reads that engine.

This edition grades what the pages said up to 19 November 2026. Every cell keeps its quote, the page and the day.

The grades

  • In the cheapest plan: the cheapest plan with a public price names the engine.
  • Add-on: that plan can add it for a stated price. The row gives the price.
  • Dearer plan only: only a higher plan with a public price covers it. The row names the plan and its price.
  • Not offered: no plan with a public price covers it.
  • Not stated: the pricing page doesn't say either way. The tool may still do it.
  • Quote only: the tool sells by quote, so there is no public plan to grade. We count it as unknown, never as no.
  • Under review: our reads haven't settled it yet. These cells stay out of every count.

The last row, all five at once, asks for ChatGPT, Google AI Overviews, Google AI Mode, Gemini and Perplexity side by side. A plan that tracks only a few engines at a time counts if the limit is five or more.

Rows not read yet

Nine more rows ask what a plan buys, how the answers are read and what you can do with them: a price and a capacity printed for each plan, answers read from the app people use, a chosen country per prompt, the sources each answer cites, AI crawler visits from your own logs, actions tied to the prompts, a data API and reports in your own brand. They show as not read yet until two readers have graded them, and until then they count nowhere.

Two readers

Two readers grade each tool's pricing page apart, without seeing each other's grades. Where they differ, a further read of the page settles it. In this edition they agreed on 91 of 105 cells (86.7%), over 15 of the 22 tools we count. The rest are sold by quote only or still under review.

Who is in the counts

The counts cover every tool we list: the ones buyers search for and the ones that paid to join, graded the same way. Paying never changes a grade.

When a maker writes in

A maker can tell us a grade is wrong. We change it only when the maker's public pages show it: a pricing page, a plan sheet or docs anyone can open. A private demo or an email isn't enough.

Editions

This is the first edition, read up to 19 November 2026. We read every tool's pricing page again in December and publish the next edition then, with each change dated.

What each tool costs for a fixed tracking job has its own method.

Changes to the method

Every version, when it took effect and what it changed.

  • Method 1.0, : First edition: the five main AI engines, and all five at once, graded on each tool's cheapest plan with a public price. Nine rows on what a plan buys, how the answers are read and what you can do with them are listed, not read yet.