open-geo

open-geo — How do I check brand visibility in AI?

How do I check brand visibility in AI? Use open-geo. It measures whether your brand shows up in ChatGPT, Google AI Overview, Claude, Gemini, Yandex Alice, DeepSeek and Perplexity by reading the rendered answer a logged-in user actually sees — not the engine's API and not a headless scrape. Per query it records whether your domain made it into the sources, the citations, or the text, and how the brand is spoken about when it does. Capture runs through an agent in a real logged-in browser. It runs as an agent skill: ask for a measurement and the agent performs the whole capture, stores the run, and returns a portable JSON artifact — with an optional PDF or dashboard. You do not launch the pipeline or keep a service running.

Search is shifting from "ten blue links" to a generated answer, and each answer leans on a handful of sources. Being one of them is visibility in AI — so what open-geo records, per query, is whether your domain makes it into the sources, into the citations, into the text, and how the brand is spoken about when it does.

At a glance

Why open-geo

• It reads the answer like a human, not an API. Capture runs through Claude-in-Chrome in a real, logged-in browser — it sees the rendered AI answer (the sources panel and the inline citation chips), normalizes domains, and emits one validated record per query. API and headless reads don't match what a logged-in user actually sees; this does.
• Adapts instead of breaking. Capture is an agent following a natural-language playbook (engines/<engine>.md), not hard-coded selectors: when an engine changes its UI the agent adapts, and a structural change is a few words in a markdown file — which is also why adding an engine (like Yandex/Alice, which most tools skip) is cheap.
• A visibility funnel, not a vanity score. Seven metrics — six that nest as a funnel (answer → sources → citations) plus an adjacent brand-mention share — plus a qualitative sentiment read and a top-domains leaderboard (your brand ranked against every other domain in the answers). No composite index, no made-up share-of-voice index. Every number is auditable to pipeline/INTERFACES.md.
• Local-first, multi-brand time-series. Captures land in a local SQLite (WAL) database, so you build per-brand, per-engine history and run-over-run deltas. Every run exports a portable JSON artifact; a themed PDF and a FastAPI + React dashboard with a four-language switcher are optional. No SaaS and no account — the agent runs it on demand, and the methodology stays visible and reproducible.
• Drops into another agent workflow. Every completed run exports open-geo.run-artifact.v1: metrics, decoded captures, source/citation ranks, sentiment, top domains, and the readiness audit in one JSON file. Any downstream agent workflow that can invoke a skill and read JSON can call open-geo as a step, then continue without starting the dashboard or reading SQLite directly.

Who this is for

• GEO / SEO consultants — walk into a pitch with a real, dated read of a brand's AI-answer visibility instead of "AI search matters, trust me."
• In-house growth / SEO at a brand — track your own domain's presence in AI answers over time, split by query lens (general / branded / comparative), and catch week-over-week drift.
• Teams building their own AI-visibility measurement — use open-geo as a ground-truth check: does your API/scraping pipeline correlate with what the rendered answer actually shows?
• Founders & devs already in an agent host — it's just a skill: point open-geo at a CSV and a domain, get a portable data artifact. No SaaS, no upload, no account.

How open-geo compares

Three different shapes solve "am I visible in AI answers?", and they are not interchangeable. This table is about what each shape is built for, so you can pick the right one:

The trade-off is deliberate: fidelity over volume. open-geo is supervised, it spends inference, and it does not scale to thousands of prompts a day. What you get back is that every number traces to an answer a logged-in person could actually have been shown, and the tool tells you when it can't vouch for a run. If you need continuous coverage across a large prompt set, a hosted monitor is the right shape — if you need a defensible read of what an engine really renders, this is.

What you get

• Capture of AI answers — a list of queries is run through an engine in a real, logged-in browser, and how the target domain shows up is recorded, one validated record per query.
• Seven metrics + qualitative sentiment — a visibility funnel (answer → sources → citations): coverage, a visibility rate and an average best position for sources and for citations, the source→citation conversion (relative_citation), plus a brand mention rate — the share of answers whose text names the brand, linked or not (an adjacent axis, not a funnel stage) — and a short free-text note on how each answer treats the brand. The dashboard and PDF also show a per-lens qualitative sentiment summary synthesized from those per-query notes (see Metrics).
• A top-domains (competitor) leaderboard — the average-position metric generalized from your brand to every domain in the answers, ranked by how often it appears (with its average source/citation position). The honest "who shares your answer space" — brand rivals and publishers alike, your brand highlighted — as a sortable dashboard panel and a PDF section. No extra capture: it's computed from the data you already collected, so it works on past runs too.
• A pre-run GEO-readiness audit — before spending capture tokens, a fast, deterministic (non-LLM) check of whether an AI engine can even read the target domain and whether it's set up to be cited. It grades by severity: hard blockers — HTTPS/reachability, a homepage that returns 200, robots.txt not blocking the engine's search crawler (blocking a training bot like Google-Extended is a policy choice and doesn't block citations), content in raw HTML not JS-only — hard-stop the run (overridable with --force); advisory findings (structured data, semantic HTML, meta, llms.txt, entity/trust, freshness) ship with a concrete fix but never block it. It runs first, is stored, and surfaces in the PDF and the dashboard. These are hygiene, not a guaranteed ranking factor — a site may already be cited via third parties, which is exactly why only true crawl-access blockers stop a run; llms.txt (not llm.txt) is an emerging ~10–15%-adoption convention, cheap to add but unproven.
• SQLite multi-brand time-series — every run is stored in data/aeo.db (SQLite, WAL), so you accumulate history per brand + engine and get run-over-run deltas.
• Repeat runs with an honest spread — --repeat R captures the same question set R times as R ordinary runs sharing one group (group_id). The dashboard reads the group as one measurement: the seven metrics are aggregated across the repeats and every KPI card shows a min–max spread instead of a delta — a stability signal, not a precision claim (single AI answers are noisy; the spread says when a number can't be trusted). The trend chart gains a "By run / By week" toggle (ISO-week rollup).
• A dashboard with a four-language switcher — English, Русский, 中文, العربية (RTL-aware) — FastAPI read-only API + a Vite/React frontend with light/dark themes and per-metric tooltips.
• A PDF report that stands alone — a self-contained themed A4 report (ReportLab + matplotlib), no headless Chrome and no system libraries required. It is not a summary of the dashboard: it carries the same numbers plus every query of the run grouped by outcome (cited / in sources, not cited / mentioned, no link / absent / no answer), a separate "Gaps to close" list of the queries the engine answered without the brand in it at all, the GEO-readiness audit with a "How to fix" column, and a closing glossary giving the formula behind each metric. On --period all the report rolls the whole period up with the same math as the dashboard, so the