agentleFS
XPOZpublicXPOZpublic/xpoz-agent-skills/skills/geo-visibility-check/SKILL.md

geo-visibility-check

Skill15 starsChanged 2 months agoMITView on GitHub

One-shot GEO visibility check for a brand using Xpoz. Runs the buyer questions that matter through Claude, ChatGPT, and Gemini, and reports where the brand appears, who wins instead, and which surfaces the answers are assembled from. Use when asked to "check my AI visibility", "does ChatGPT recommend us", "GEO audit", "where do we show up in AI answers", or "who wins our category in AI answers".

  • Reads credentials
  • Installs packages

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 7 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before fetching data, ensure Xpoz access is configured.

    line 25
  2. Distinctive:

    If this succeeds without error → skip to Step 1.

    line 36
  3. Distinctive:

    If neither works, you need to authenticate.

    line 38
  4. Very specific:

    If mcporter is available:

    line 50
  5. Very specific:

    If hasAccessKey: true → ready.

    line 56
  6. Distinctive:

    Step 3: WAIT for the user to reply with the code: Do not proceed until they respond.

    line 117
  7. Very specific:

    Run each panel prompt through the AI engines with citation capture.

    line 240
  8. Distinctive:

    No engine keys at all? Deliver the demand-side half only: the panel itself, built from real community phrasings, plus where those questions …

    line 244
  9. Distinctive:

    Keep the trace JSONs and the report.

    line 298
  10. Distinctive:

    If community citations dominate the lost prompts, run geo-reddit next to map the exact venues.

    line 299

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

GEO Visibility Check

Overview

Buyers increasingly start, and often finish, their research inside AI engines. This skill answers the first question of any GEO effort: for the questions that matter, does the brand appear in AI answers, and if not, who does and through which sources? It assembles a small panel of buyer prompts, runs each through the engines with full citation tracing, and delivers a per-prompt visibility verdict plus a read on where the winning answers come from. One-shot by design; it is the measurement pass that tells you whether a full GEO program is worth running.

When to Use

Activate when the user asks:

  • "Do AI assistants recommend [BRAND]?"
  • "Check our AI/GEO visibility"
  • "Where does [BRAND] show up in ChatGPT/Claude/Gemini answers?"
  • "Who wins [CATEGORY] in AI answers, and why?"
  • "Run a GEO audit for [PRODUCT]"
  • "Which sources do AI engines use for our category?"

Setup & Authentication

Before fetching data, ensure Xpoz access is configured. Follow these checks in order.

Check 1: Already authenticated?

If you have MCP tools, try calling any Xpoz tool (e.g., checkAccessKeyStatus). If it works → skip to Step 1.

If you have the SDK, try:

from xpoz import XpozClient
client = XpozClient()  # reads XPOZ_API_KEY env var

If this succeeds without error → skip to Step 1.

If neither works, you need to authenticate. Get a free access key (below).


Recommended: a free access key

Real analyses need a real key: get a free access key (free tier, up to 75K results, no credit card). SDK and CLI users set it as XPOZ_API_KEY; MCP connections sign in with the same account via OAuth on first tool call (paths below).


Path A: MCP via mcporter (OpenClaw agents)

If mcporter is available:

mcporter call xpoz.checkAccessKeyStatus

If hasAccessKey: true → ready. If not:

mcporter config add xpoz https://mcp.xpoz.ai/mcp --auth oauth

Then authenticate — generate the OAuth URL and send it to the user:

Step 1: Generate authorization URL

import secrets, hashlib, base64, urllib.parse, json, urllib.request, os

verifier = secrets.token_urlsafe(64)
challenge = base64.urlsafe_b64encode(hashlib.sha256(verifier.encode()).digest()).rstrip(b'=').decode()
state = secrets.token_urlsafe(32)

# Dynamic client registration
reg_req = urllib.request.Request(
    'https://mcp.xpoz.ai/oauth/register',
    data=json.dumps({
        'client_name': 'Agent Skills',
        'redirect_uris': ['https://www.xpoz.ai/oauth/openclaw'],
        'grant_types': ['authorization_code'],
        'response_types': ['code'],
        'token_endpoint_auth_method': 'none',
    }).encode(),
    headers={'Content-Type': 'application/json'},
)
reg_resp = json.loads(urllib.request.urlopen(reg_req).read())

params = urllib.parse.urlencode({
    'response_type': 'code',
    'client_id': reg_resp['client_id'],
    'code_challenge': challenge,
    'code_challenge_method': 'S256',
    'redirect_uri': 'https://www.xpoz.ai/oauth/openclaw',
    'state': state,
    'scope': 'mcp:tools',
    'resource': 'https://mcp.xpoz.ai/',
})

auth_url = 'https://mcp.xpoz.ai/oauth/authorize?' + params

# Save state for token exchange
os.makedirs(os.path.expanduser('~/.cache/xpoz-oauth'), exist_ok=True)
with open(os.path.expanduser('~/.cache/xpoz-oauth/state.json'), 'w') as f:
    json.dump({'verifier': verifier, 'state': state, 'client_id': reg_resp['client_id'],
               'redirect_uri': 'https://www.xpoz.ai/oauth/openclaw'}, f)

print(auth_url)

Step 2: Send the URL to the user

Tell them:

"I need to connect to Xpoz for social media data. Please open this link and sign in:

[auth_url]

After authorizing, you'll see a code. Paste it back to me here."

Step 3: WAIT for the user to reply with the code. Do not proceed until they respond.

Step 4: Exchange the code for a token

Once the user provides the code (either a raw code or a URL containing ?code=...), extract the code and exchange it:

import json, urllib.request, urllib.parse, subprocess, os

with open(os.path.expanduser('~/.cache/xpoz-oauth/state.json')) as f:
    oauth = json.load(f)

code = "THE_CODE_FROM_USER"  # Extract from user's reply

data = urllib.parse.urlencode({
    'grant_type': 'authorization_code',
    'code': code,
    'redirect_uri': oauth['redirect_uri'],
    'client_id': oauth['client_id'],
    'code_verifier': oauth['verifier'],
}).encode()

req = urllib.request.Request(
    'https://mcp.xpoz.ai/oauth/token',
    data=data,
    headers={'Content-Type': 'application/x-www-form-urlencoded'},
)
resp = json.loads(urllib.request.urlopen(req).read())
token = resp['access_token']

# Configure mcporter with the token (token is never printed)
subprocess.run(['mcporter', 'config', 'remove', 'xpoz'], capture_output=True)
subprocess.run(['mcporter', 'config', 'add', 'xpoz', 'https://mcp.xpoz.ai/mcp',
                '--header', f'Authorization=Bearer {token}'], check=True)

# Clean up
os.remove(os.path.expanduser('~/.cache/xpoz-oauth/state.json'))
print("Xpoz configured successfully")

Step 5: Verify with mcporter call xpoz.checkAccessKeyStatus → should return hasAccessKey: true.


Path B: MCP via Claude Code

For Claude Code users without mcporter:

claude mcp add --transport http xpoz https://mcp.xpoz.ai/mcp

Claude Code handles OAuth automatically on first tool call — the user just needs to authorize in their browser when prompted.


Path C: SDK (Python or TypeScript)

Ask the user:

"I need a Xpoz API key to access social media data. Please go to https://xpoz.ai/get-token (it's free, no credit card needed) and paste the key back to me."

WAIT for the user to reply with the key. Then:

Python:

pip install xpoz
from xpoz import XpozClient
client = XpozClient("THE_KEY_FROM_USER")

TypeScript:

npm install @xpoz/xpoz
import { XpozClient } from "@xpoz/xpoz";
const client = new XpozClient({ apiKey: "THE_KEY_FROM_USER" });
await client.connect();

Or set the environment variable and use the default constructor:

export XPOZ_API_KEY=THE_KEY_FROM_USER

Auth Errors

ProblemSolution
MCP: "Unauthorized"Re-run the OAuth flow above
SDK: AuthenticationErrorVerify key at xpoz.ai/settings
Token exchange failsAsk user to re-authorize — codes are single-use

Step-by-Step Instructions

Step 1: Build the Prompt Panel

Collect 5-10 buyer prompts. Sources, best first:

  1. The user's own list: the questions they believe buyers ask. Take these as given.
  2. Community demand via Xpoz: mine the real phrasings buyers use. Search for questions around the problems the product solves:
Call getRedditPostsByKeywords:
  query: "<problem phrasings, OR-joined, e.g. 'best tool for X' OR 'how do I X' OR 'X alternative'>"
  fields: ["id", "title", "subredditName", "score", "commentsCount", "createdAtDate"]
  limit: 15
  startDate: "<90 days ago, YYYY-MM-DD>"
  endDate: "<today, YYYY-MM-DD>"

The default fast mode returns results directly (pass limit; queries cap at 250 characters); only responseType: "paging"/"csv" calls return an operationId to poll via checkOperationStatus (every ~5 seconds until finished). Repeat with getTwitterPostsByKeywords (pass filterOutRetweets: true) for the developer/founder conversation.

  1. Derived: phrase the product's jobs-to-be-done as assistant questions ("best [category] for [persona]", "how do I [job]", "[current tool] alternatives", "how much does [category thing] cost").

Balance the panel across intents: category ("best X"), comparison ("X vs Y", "X alternatives"), problem-solution ("how do I..."), and pricing. Confirm the panel with the user before spending engine calls on it.

Step 2: Trace Every Prompt

Run each panel prompt through the AI engines with citation capture. Use the ai-answer-trace skill if it is installed. Otherwise fetch https://raw.githubusercontent.com/XPOZpublic/xpoz-agent-skills/main/skills/ai-answer-trace/SKILL.md and follow it. If neither is possible, use the degradation ladder that skill describes.

Defaults: every engine with a key configured, 2 samples per engine per prompt. That is prompts x engines x 2 paid API calls; state the number and get a nod before running a large panel.

No engine keys at all? Deliver the demand-side half only: the panel itself, built from real community phrasings, plus where those questions concentrate. Label the report clearly as demand research without engine measurement.

Step 3: Judge Visibility Per Prompt

For each prompt and engine, from the trace JSONs:

  • Present: is the brand named in the answer? How: recommended outright, listed among options, mentioned in passing, or absent?
  • Cited: does any brand-owned URL appear in cited_urls (site, docs, listings)? An answer can name the brand while citing someone else's page about it; record the actual citation path.
  • Winners: who is recommended instead or above, and via which cited URLs?
  • Stability: consistent across samples, or a coin flip?

Aggregate cited domains across all traces (the cited-domains.py chunk in ai-answer-trace) and classify the top domains by surface type: own sites, review sites and directories, community threads, documentation, publisher listicles.

Step 4: Generate Report

## GEO Visibility Check: [BRAND]
**Panel:** [N] prompts | **Engines:** [list] | **Samples:** [N] per engine | **Date:** [date]

### Verdict
[2-3 sentences: overall visibility, the strongest and weakest surface, the single most valuable lost prompt]

### Per-Prompt Visibility
| Prompt | Claude | ChatGPT | Gemini | Who wins |
|--------|--------|---------|--------|----------|
| [prompt] | [recommended / listed / absent] | ... | ... | [competitor + citation path] |

### Where the Answers Come From
| Domain | Instances | Surface type |
|--------|-----------|--------------|
| [domain] | [n] | [own site / review site / community / docs / listicle] |

### Brand Citation Paths
[For each prompt where the brand appears: which URL carried it into the answer]

### Lost Prompts, Diagnosed
[For each absent/losing prompt: what the winning citations are, and what kind of asset could enter that surface]

### Recommended Next Steps
[3-5 bullets tied to the evidence: which surface to work first, which prompts are winnable]

Example Prompts

  • "Check whether AI assistants recommend xpoz for social media data"
  • "GEO audit for [BRAND]: 8 prompts, all three engines"
  • "Does ChatGPT know about our product? Who does it recommend for [CATEGORY]?"
  • "Which sources do AI engines assemble [CATEGORY] answers from?"
  • "Find the buyer questions where we're absent from AI answers"

Notes

  • Step 2 needs AI engine API keys (see ai-answer-trace); Step 1's community mining needs only xpoz.
  • Engine answers are nondeterministic; the per-prompt verdict needs 2+ samples to mean anything.
  • Keep the trace JSONs and the report; a re-run next month against the same panel is a real before/after.
  • If community citations dominate the lost prompts, run geo-reddit next to map the exact venues.
  • Free access key: up to 75K results at xpoz.ai (no credit card); real runs need it
  • For the weekly program this check previews (tracked prompt panel, dated snapshots, gap analysis, content production and rework), use geo-seo-agent.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.