An AI analyst that answers from your own data
Ask why visibility moved, where competitors win or what to fix next. Agent M checks your prompts, citations, audits and connected Google data with 50 tools, answers with charts, and asks before it runs a live scan.
Product preview, illustrative data
What it does
Answers built on tool calls
Agent M answers by calling tools over your tracked prompts, citations, competitors, audits and connected Search Console and GA4. Each call shows as a chip you can open to inspect what it returned.
Charts and tables in the chat
Ask for a trend or a comparison and it draws a card with metric tiles, line or bar charts, tables and lists from the tool results in that conversation. Up to three buttons send the obvious follow-up question.
Live scans wait for your yes
A live scan never runs without your approval. Agent M shows the estimated credits and your balance before it asks, and a scan given that figure as its ceiling refuses to cost more. The first new tracked prompt in a session asks too.
Memory the whole team shares
It reads the project's memory before planning work: business overview, ideal customer, positioning, goal, key pages and a research log. It writes what it learns back, and Settings marks whether a person, Agent M or onboarding last updated each section.
Drop triage with evidence
Ask why traffic or AI visibility fell and it dates the change, checks whether impressions held while clicks dropped, and ranks likely causes by evidence. It also says what the data cannot rule out.
Reports your team can open
Ask for a report or a client check-in and it saves one self-contained HTML page to Reports, updating an earlier version rather than duplicating it. A workspace manager can share a read-only link.
Three steps, start to finish
Step 1: Ask in plain language
Open Agent M from the sidebar and ask about visibility, citations, competitors or a page. Starter questions cover a drop, where competitors win, pages to optimize and a summary of the week.
Step 2: Watch it work
It reads project memory, calls the tools it needs and shows each call as it runs. Longer jobs follow built-in workflows: a visibility audit, closing citation gaps, a report or a prompt review.
Step 3: Approve, save, follow up
Approve a live scan from the card in the chat, save the findings as a report, and let Agent M log the verdict to the research log so the next question starts from it.
The specifics
Engines, data sources, cadence and plan limits, stated as the product works today.
- Where it lives
- Agent M in the app sidebar. Chats save automatically to a per-project history that only you can see
- Model and tools
- Claude Sonnet 4.6 with 50 tools, 43 of them shared with the Measure MCP server. Live scans, prompt tracking and in-chat charts are Agent M only
- Data it reads
- Your tracked-prompt scans, citations, competitors, audits, accuracy checks and reports, a market-wide corpus of Google AI Overview and ChatGPT answers (US, English), plus Search Console and GA4 once connected
- Messages per day
- Starter 25, Pro 75, Advanced 200, Enterprise 500. Shared across the workspace, enforced on the server, reset at 00:00 UTC
- Approvals and credits
- Live scans ask every time, a new tracked prompt the first time in a session. Live scan credits per engine: Perplexity 1, AI Overview 2, ChatGPT, Gemini and Claude 3
- Limits
- It cannot run code, browse files or edit your website, and it only works when you send it a message. No scheduled runs
Common questions
What data does Agent M use to answer questions?
Agent M answers from tool calls over your Measure workspace: your tracked-prompt scans (ChatGPT, Perplexity, Gemini and Claude), citations, competitors, site audits, AI accuracy checks and saved reports. It can also read a market-wide corpus of Google AI Overview and ChatGPT answers (US, English) and, once connected, your Search Console and GA4 data. Every tool call appears in the chat, and you can open it to inspect the result.
Can Agent M spend credits without asking?
A live scan never runs without your approval. Agent M shows the estimated credits and your balance before it asks, and when it passes that figure as a ceiling the scan refuses to run if the real cost would be higher. Engine calls that fail without being billed are refunded. Some data lookups, such as LLM Mentions, the visibility history or backlinks, cost the same credits as in the dashboard when the data is not already cached, and those do not show an approval card.
How many Agent M messages do I get per day?
Starter includes 25 messages a day, Pro 75, Advanced 200 and Enterprise 500. The allowance is shared by everyone in the workspace, checked on the server before any model work starts, and resets at 00:00 UTC. When it runs out, the chat box locks and shows an upgrade option until the next reset.
Does Agent M remember earlier conversations?
Each chat is saved automatically to a per-project history that only you can see, and you can reopen it later. Durable knowledge lives in project memory instead: business overview, ideal customer, positioning, current goal, writing preferences, key pages and a research log. Agent M reads it before planning work and writes back what it learns, and you can edit the same record in Settings.
Can Agent M make changes on its own?
No. Agent M only works when you send it a message, and it never runs on a schedule. It cannot run shell commands, browse files or edit your website. Inside Measure it writes tracked prompts, saved audits and reports, AI accuracy verdicts, brand facts and project memory, and it spends workspace credits on live scans and uncached lookups. Live scans and the first new prompt in a session wait for your approval.
Know where your brand stands in AI answers.
Plans start at $89 a month. Add your brand, choose the prompts your buyers ask, and see who AI engines recommend.