Free tools and skills for Claude and other AI assistants, built on the same method the Census uses. Most GEO tools give you a score. These give you counts you can trace back to data, say how sure the data lets you be, and grade every tactic by the evidence behind it. They are licensed CC BY 4.0, so you can use and adapt them, including for client work.
Use it in your browser
Content Shape Checker. Paste a draft and see whether it is built like human writing or generated text. It runs four tested checks (over-explanation, named specifics, rhythm and variety, and stock AI phrasing), with every flagged sentence quoted and a fix for each. Free, no signup, and your text never leaves your browser.
Skills for Claude and other assistants
| Skill | What it does | You need |
|---|---|---|
| Question set builder | Builds a fixed set of buyer questions with stable IDs, spread across four intents (vendor selection, comparison, specific need, educational), so results can be compared month to month. | Nothing |
| Answer log analyzer | Reads logged AI answers from repeated runs and reports how often a brand is named, cited, cited while a competitor is named, or absent. Shows the run-to-run spread, a 95% interval for each rate, and whether a month-on-month change is outside the noise. | A CSV of answers from a tracker or logged by hand |
| Schema parity checker | Checks that structured data says only what the visible page says: FAQ text, prices, availability, ratings, phone numbers, dates and headlines. | A URL or a saved HTML file |
| GEO migration guard | Protects AI search visibility through a site move: records a baseline, checks the redirect map with cited URLs first, compares pages after launch, and flags AI crawlers newly blocked in robots.txt. | URL lists and a redirect map |
| AI crawler log check | Finds AI and search crawlers in server logs and verifies them against the IP ranges the companies publish, or by reverse DNS, so spoofed bots are not counted. | Server or CDN access logs |
| Content shape checker | A version of the four checks in the web tool to run inside Claude, with every flagged sentence quoted and a fix for each. | A draft with headings |
| GEO experiment kit | Templates and a small script for testing one change at a time with control pages, written predictions and an honest readout, including results that show nothing. | Nothing |
The scripts are Python 3 with no extra packages to install (the schema checker can optionally render pages with Playwright). Each skill comes with a worked example built from made-up data, showing the real output.
Evidence grades
Every tactic a skill recommends is tagged with the strength of the evidence behind it: tested by the Census, a published study, the platform’s own documentation, a vendor claim, or untested. The grades and their sources are in one file, and they change only when the evidence does. Results from our own tests are on the experiments page.
A typical workflow
- Build the questions with the question set builder.
- Ask them on each AI engine, at least three times per check date, and log the answers.
- Run the answer log analyzer on the log.
- Fix what the data points to, one change at a time, with the experiment kit.
- Use the schema checker, crawler log check and migration guard for the technical side.
How to install
- Claude (web or desktop): zip the skill folder, then in Settings > Capabilities turn on code execution, scroll to Skills, click “Upload skill” and choose the zip. Anthropic’s guide has the details.
- Claude Code: copy the skill folder into
~/.claude/skills/for personal use, or.claude/skills/inside a project. - Other assistants: each skill is a plain Markdown file. Paste it into a custom GPT, a Gem or a project’s instructions. The scripts also run on their own from the command line.
What these skills will not do
They do not promise that a business will be named by an AI tool, they do not invent visibility scores, and they contain no hidden instructions or tricks aimed at AI systems. They follow the same rules as the rest of this site, set out in our editorial standards.
Feedback
Found a problem, or think a grade is wrong? Open an issue on GitHub or email [email protected].