Short answer: Muse is Meta’s AI agent app, built on the same Meta model as Meta AI. You can’t see inside the app’s own actions, but you can measure what Meta AI answers when your buyers ask about your category: whether your brand is named, where it comes in the answer, which rivals appear beside it, and whether your pages are cited. Our Muse / Meta AI visibility API returns those answers from Meta AI as JSON, at one credit per answer ($1.99 per 1,000 pay-as-you-go), and the first 30 requests are free. The script below tracks a prompt set weekly and shows what changed.

What Muse is

Muse is Meta’s personal AI agent app, launched in the US on 8 September 2026. Meta describes it as an agent that “doesn’t just answer questions, it actually does the work”: it can send emails, book travel, fill out forms and, with your approval, make purchases (Meta, 8 September 2026).

It took off quickly. In September 2026 it reached #1 on the US App Store’s free iPhone chart, displacing ChatGPT (Appfigures, 20 September 2026; TechCrunch, 21 September 2026). By 20 October it had 10.9 million downloads worldwide, 10.2 million of them in the US, according to Sensor Tower data reported by TechCrunch. We checked each source on 18 November 2026.

Muse and Meta AI share a model

Meta says Muse is “powered by Muse Spark, Meta’s most capable model to date” (Meta, 8 September 2026), and that “Muse Spark now powers the Meta AI assistant in the Meta AI app and meta.ai” (Meta, 8 April 2026). So the closest thing you can measure to what Muse tells buyers is what Meta AI tells them: the assistant built on the same model.

We return answers from Meta AI. We don’t return the Muse app’s own screens or answers, and we don’t track anything the app does on a buyer’s behalf, such as shopping, browsing or buying.

What you can measure, and what you can’t

You can measureYou can’t measure
Whether Meta AI names your brand for a buyer’s promptWhat the Muse app does for a buyer: browsing, shopping, buying
Where it first names you, against your rivalsWho asked, or how often a prompt is asked
How many times it mentions you, with the sentence around each mentionA buyer’s own follow-up conversation
Whether it cites your pages, and which sources it cites
Which other brands it names

The answers change from run to run, so treat one answer as one data point and track the same prompts over time.

A tested weekly workflow

  1. Write the prompts your buyers would type. Use their words, not your keywords: “What is the best running shoe for flat feet?” rather than “flat feet running shoe”.
  2. List your brand and your rivals. Give your brand its domain, so the API can tell when Meta AI cites your pages, not just your name.
  3. Ask Meta AI each prompt every week and append the result to a CSV.
  4. Compare each run with the previous one and look at what moved.

This script does steps 3 and 4. It asks Meta AI each prompt through the /chatgpt/mentions endpoint with provider: "meta_ai", records whether your brand was named, its order, whether it was cited and which rivals appeared, and prints what changed since the last run:

import csv
import os
from datetime import datetime, timezone
from pathlib import Path

import requests

BRAND = {"name": "Brooks", "domains": ["brooksrunning.com"]}
COMPETITORS = ["Hoka", "ASICS", "Saucony", "New Balance"]
PROMPTS = [
    "What is the best running shoe for flat feet?",
    "Brooks or Hoka for overpronation?",
    "Most comfortable running shoes for beginners",
]
OUT = Path("meta-ai-visibility.csv")
FIELDS = ["run", "prompt", "mentioned", "order", "mentions", "cited", "competitors_named"]
HEADERS = {"Authorization": f"Bearer {os.environ['FETCHLAYER_API_KEY']}"}

run = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M")
rows = []
for prompt in PROMPTS:
    res = requests.post(
        "https://api.fetchlayer.dev/chatgpt/mentions",
        json={"prompt": prompt, "provider": "meta_ai", "targets": [BRAND, *COMPETITORS], "includeAnswer": False},
        headers=HEADERS,
        timeout=240,
    )
    res.raise_for_status()
    targets = {t["name"]: t for t in res.json()["targets"]}
    me = targets[BRAND["name"]]
    rows.append({
        "run": run,
        "prompt": prompt,
        "mentioned": me["mentioned"],
        "order": me["orderOfFirstMention"] if me["mentioned"] else "",
        "mentions": me["mentionCount"],
        "cited": me["citedAsSource"],
        "competitors_named": "; ".join(c for c in COMPETITORS if targets[c]["mentioned"]),
    })

new_file = not OUT.exists()
with OUT.open("a", newline="", encoding="utf-8") as f:
    writer = csv.DictWriter(f, fieldnames=FIELDS)
    if new_file:
        writer.writeheader()
    writer.writerows(rows)

# Compare with the previous run in the file
with OUT.open(encoding="utf-8") as f:
    history = list(csv.DictReader(f))
runs = sorted({r["run"] for r in history})
if len(runs) > 1:
    before = {r["prompt"]: r for r in history if r["run"] == runs[-2]}
    for r in history:
        if r["run"] != runs[-1] or r["prompt"] not in before:
            continue
        b = before[r["prompt"]]
        changes = [f"{k}: {b[k] or '-'} -> {r[k] or '-'}" for k in ("mentioned", "order", "cited", "competitors_named") if b[k] != r[k]]
        print(f"{r['prompt'][:45]:<45} {'; '.join(changes) or 'no change'}")
else:
    for r in rows:
        print(f"{r['prompt'][:45]:<45} named={r['mentioned']} order={r['order'] or '-'} cited={r['cited']} rivals={r['competitors_named']}")

The first run on 18 November 2026 (3 prompts, 3 credits):

What is the best running shoe for flat feet?  named=True order=1 cited=False rivals=Hoka; ASICS; Saucony; New Balance
Brooks or Hoka for overpronation?             named=True order=1 cited=False rivals=Hoka
Most comfortable running shoes for beginners  named=True order=1 cited=False rivals=Hoka; ASICS; Saucony; New Balance

And the second run, a few minutes later:

What is the best running shoe for flat feet?  no change
Brooks or Hoka for overpronation?             no change
Most comfortable running shoes for beginners  no change

Brooks was named first in all three answers, but never cited: Meta AI named it without linking to brooksrunning.com. That’s the kind of gap worth working on, and only a run that checks citations shows it. The diff ignores raw mention counts, which move a little between runs even when nothing that matters has changed.

Run it weekly with cron, a GitHub Actions schedule or any scheduler you already use. Five prompts a week is about 20 answers a month, roughly $0.04 pay-as-you-go.

Without code

Connect FetchLayer’s MCP server to Claude, ChatGPT or Cursor (setup) and ask:

Ask Meta AI “What is the best running shoe for flat feet?” and tell me whether Brooks is named, in what order against Hoka, ASICS and Saucony, and which sources it cites.

Next steps