# Citelens Check

> Run a live GEO (Generative Engine Optimization) visibility check on a brand from the terminal — measure how often ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews name or cite a brand vs. its competitors. Use when the user says "check my GEO", "run a citelens check", "how visible is <brand> in AI answers", "do AI assistants recommend <brand>", or "share of voice in AI search".

- Skill: `manavarya09/citelens-check` (Agent Skill)
- Install (CLI): `npx skillmds@latest add manavarya09/citelens-check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/manavarya09/citelens-check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: manavarya09 (https://skillmd.com/u/manavarya09)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/manavarya09/citelens-check

---


# Citelens GEO check

Run the Citelens GEO pipeline (buyer-intent prompt generation → query each AI
engine → parse the answers → compute visibility metrics) against a brand and
print a terminal report. This is the same pipeline the Citelens web app uses,
run standalone with no database.

## Inputs you need

Gather these from the user's request. Only `brand` is required — infer or ask
for the rest.

- **brand** (required): the brand/product name to measure, e.g. `Linear`.
- **domain**: the brand's domain, e.g. `linear.app`. Used to detect when the AI
  cites the brand's *own* site. If unknown, pass empty.
- **topics**: the buyer categories to probe, comma-separated, e.g.
  `project management,issue tracking`. These drive prompt generation — the brand
  name is deliberately never injected, so the check measures whether engines
  bring the brand up *unprompted*. Defaults to the brand name if omitted.
- **competitors**: comma-separated rivals to compare share-of-voice against,
  e.g. `Jira,Asana,Notion`.

If the user gave only a brand, infer plausible topics and 2–4 well-known
competitors from what the brand does, state your inferences, and proceed. Ask
via AskUserQuestion only if the brand is ambiguous.

## How to run

The runner lives in the Citelens repo at `scripts/geo-check.ts`. Run it from the
repo root with `tsx` (already a dev dependency):

```bash
npx tsx scripts/geo-check.ts \
  --brand "Linear" \
  --domain linear.app \
  --topics "project management,issue tracking" \
  --competitors "Jira,Asana,Notion"
```

Add `--engines gemini` to hit a single engine, or `--json` for machine-readable
output you can summarize.

## Live vs. mock

- With **no API keys**, every engine falls back to a deterministic mock so the
  command always works offline — but a niche brand will read 0% visibility,
  which is the expected "the AI never brings you up" signal, not a bug.
- For **real data**, the user sets `GEMINI_API_KEY` (free from
  https://aistudio.google.com/apikey) — Gemini's Google Search grounding returns
  real cited sources. Other engines go live when their key is set
  (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `PERPLEXITY_API_KEY`). See
  `.env.example`.
- For **no-key real UIs**, `CITELENS_BROWSER=1` drives the real Perplexity /
  Google AI Overview pages (experimental). Mention this only if the user asks to
  avoid API keys.

The report footer prints how many calls were live and the estimated cost. If
0 live calls, tell the user the numbers are from the mock and how to go live.

## Reading the result for the user

After running, summarize in plain language: the visibility rate (is the brand
named or cited at all?), where it sits in share-of-voice vs. competitors, which
engine is weakest, and what kind of sources the engines cited (review sites,
forums, media). If visibility is low, point out that the full Citelens app
closes the loop — it diagnoses *why* (who got cited instead, from what source)
and drafts the fix. Suggest `/citelens:setup` to launch the dashboard for that.