Monitor your AI share of voice

AI share of voice is one ratio: of all the times an AI answer names a brand in your category, how many of those times is it you. It is the AI-answer era's version of the rank report — except there is no public rank to read, so the only honest way to get the number is to fetch the answers themselves and count. AgentGEO does the fetching: one API call returns the answer ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode or Copilot gave, as JSON. The counting stays in your code — twenty lines, shown below — so the number you report is one you can audit down to the individual answer.

The job: a recurring number per engine — your mentions divided by all tracked-brand mentions — over a fixed query set, with history you can diff. Everything else on this page is the shortest honest path to that.

How share-of-voice monitoring works on raw answers

1. Fix a query set. Ten to fifty questions your buyers actually ask — "best [category] for [audience]", "[competitor] alternatives", "how do I [job]". The set must stay fixed between runs, or the trend line measures your edits, not the market.
2. Fetch every query against every engine you care about. One POST /v1/fetches per query names the engines in surfaces; each delivered answer is one record, one credit.
3. Count mentions and divide. Case-insensitive search of answerText for each tracked brand, yours included. Share of voice = your mentions ÷ all tracked-brand mentions. Store the per-answer rows, not just the ratio, so any number can be traced back to the answer that produced it.

import requests

BRANDS = ["Acme", "Northwind", "Globex"]  # yours first
QUERIES = [
    "best project management software for agencies",
    "project management tools with client portals",
]
SURFACES = ["chatgpt", "perplexity", "gemini"]

mentions = {b: 0 for b in BRANDS}
for query in QUERIES:
    resp = requests.post(
        "https://api.agentgeo.org/v1/fetches",
        json={"query": query, "surfaces": SURFACES},
        headers={"Authorization": "Bearer ag_live_your_key_here"},
        timeout=200,  # the API holds up to 180s on slow surfaces
    )
    resp.raise_for_status()
    for answer in resp.json()["answers"]:
        text = (answer.get("answerText") or "").lower()
        for brand in BRANDS:
            if brand.lower() in text:
                mentions[brand] += 1

total = sum(mentions.values()) or 1
for brand, n in sorted(mentions.items(), key=lambda kv: -kv[1]):
    print(f"{brand:10} {n:3} mentions  {n / total:6.1%} share of voice")

Run it weekly — a cron, a CI job, or a schedule in the AgentGEO console — and append each run's rows to a file. The diff between two runs is the deliverable: who gained, who lost, on which questions, on which engines.

Let your agent run it

The same data is exposed over MCP, and the geo-share-of-voice skill — one of eight that ship inside the agentgeo-mcp package — walks any connected agent through exactly this workflow: fetch, count, compare, report. Install once and ask in plain language:

claude mcp add agentgeo -- npx -y agentgeo-mcp --key ag_live_...

Alternatives

GEO dashboards like Profound or Peec AI will hand you a share-of-voice score. The trade: their query set, their counting rules, their seat pricing — and a number you can't decompose when a client asks where it came from. Scraping the engines yourself gets you raw data and a permanent browser-automation maintenance bill. The data-layer position is the middle: raw records from one API, counting rules you wrote, a number you can defend.

More use cases

The same records power competitor visibility tracking and AI Overviews monitoring — see all use cases.