Potato Review

Overview

Potato measures how visible your brand is inside Claude's web-search answers, which is a specific corner of the AI search optimization (GEO) problem. It asks Claude a fixed set of frozen questions, collects every brand mention and source citation in the answers, and scores them with deterministic rules. No AI judge, no guesswork. Every number in the report carries a confidence interval, and the whole thing runs locally so results are reproducible and auditable. The output is a single-file offline HTML report covering mention coverage, citation validity, and the split between citations from your own domain versus third-party pages. The report is honest about its limits: it measures one engine, Claude, under one exact configuration, and it says so in the output. It is a proxy measurement, not a ranking truth detector. A free mock preview mode runs with no API key so you can see the report shape before spending anything. Setup is the friendliest of any tool in this batch. Windows users download a portable zip and double-click a batch file. Developers can pip install the package and run the CLI or the local GUI wizard. Real runs against Claude need your own Anthropic API key, which is also the only cost. The closest directory entry is Claude SEO, which audits your whole site for citability. Potato is narrower and complementary: it measures what Claude actually says about you today, repeatedly, so you can track whether fixes move the numbers. It fits brands that care specifically about Claude citations, analysts who want reproducible measurement, and teams that refuse to send brand data to a third-party monitoring SaaS. If you need cross-engine coverage of ChatGPT, Gemini, and Perplexity too, the commercial AI visibility platforms in this category are the broader option.

AI Capabilities

• Measures brand mention coverage in Claude web-search answers
• Citation validity checking (are cited links reachable)
• Owned vs earned citation split
• Deterministic scoring with confidence intervals, no AI judge
• Offline single-file HTML report

Key Integrations

Hands-on notes

The README's own framing is refreshingly honest: not a crawler, not an AI-ranking truth detector, a reproducible proxy measurement of how visible your brand is inside Claude's web-search answers, with a fixed set of frozen questions, deterministic scoring rules, no AI judge, and confidence intervals on every number. Distribution is unusually accessible: a Windows zip with no Python needed, plus a standard Python install for everyone else. 254 passing tests are claimed in the repo. Everything here comes from the repository documentation; we have not run a visibility measurement with it.

The cost model is clean: the tool is MIT with zero author fees, a free mock preview needs no API key, and real runs use your own Anthropic key, hard-capped to a budget you set, typically around $5 or less on the Haiku tier, with the estimate shown before it starts. Security posture is documented concretely: runs on 127.0.0.1, the key stays in memory, zero telemetry. The limitation is scope: it measures Claude answers only, under a fixed question set, so treat results as one engine's proxy signal rather than an industry-wide AI-visibility metric.

Verdict

The most methodologically honest AI-visibility tool in this directory: scoped claims, deterministic scoring, cost-capped runs, and a reproducible method. Use it to track your Claude-answer presence over time; do not mistake it for a full AI-search measurement.

Similar Tools