Track Your Brand in AI Answers With a Weekly Agent Summary

The short answer: If someone on the leadership team has started asking whether ChatGPT recommends you, a fixed list of buyer questions tracked daily and a Monday summary checked by one person will answer them. SE Ranking's Core plan tracks 100 prompts a day and includes MCP access for $129 a month billed monthly (SE Ranking pricing), and Make's Free plan carries 1,000 credits a month for the weekly agent run (Make pricing), both checked on 26 October 2026. For example, asking 30 questions in 3 assistants by hand is 30 x 3 = 90 answers to read and log each week; with the agent, the reviewer reads one page and opens only the answers that changed.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

An AI answer is not a list of ten links. It is a paragraph that names some brands, leaves others out and cites a handful of sources. Tracking it means asking the same questions on a schedule and keeping what came back, so that any change has something to be compared with.

What decides the setup:

1. Size the list to the allowance. SE Ranking's 14-day trial tracks 20 AI prompts a day and Core 100 (SE Ranking pricing); Nightwatch Starter allows 50 prompts and 1,500 collected answers a month (Nightwatch pricing). Twenty to thirty questions is a sound start, and well inside either.
2. Take the questions from people who already ask them. Sales call notes, support tickets and your Search Console queries hold the wording buyers use. Write each prompt the way a person would type it into a chat, then name your brand and the three rivals you lose deals to, so each answer has a field to compare against.
3. Fix the number before the first run. Report the share of tracked prompts whose answer names you, per assistant, and list the sources cited when you are missing. Collect four weeks of it before anyone changes a page.

The workflow: daily collection, one summary a week

1. The tracker collects every day. In Nightwatch, prompts go under LLM Tracking and your brand and rivals under Brand Entities; each answer is stored with the brands detected, their position, sentiment and the domains cited, refreshed daily (LLM tracking guide). In SE Ranking, prompts sit in the project beside your keywords. The set-up screens for each are in how to use Nightwatch and how to use SE Ranking.
2. Trigger. A Make scenario scheduled Weekly, early on Monday (scheduling).
3. The agent reads the tracker. Make's MCP Client app, an open beta, has an Execute an action with AI module that picks and runs tools on a remote MCP server from a task you describe. For Nightwatch, enter https://mcp.nightwatch.io/sse by hand and paste a scoped API key into the API Key / Access token field; Nightwatch keys are read-only unless you opt into write scopes (Nightwatch MCP). SE Ranking's server is https://api.seranking.com/mcp and signs in by OAuth or an X-Api-Key header, and its keys carry full account access because read-only keys are not available yet (SE Ranking MCP). In the module's Tools field, select only the reading tools. Make recommends testing a server with the Call a tool module first, so run one call by hand before you schedule anything.
4. The agent writes against a fixed brief. The Task field says: compare this week with last week; list the prompts where the brand appeared or dropped out, per assistant; name the brands that took its place; list newly cited domains; flag any change in sentiment; quote the prompt behind every line, and say nothing the data does not show.
5. One person checks before anyone else reads it. A Gmail module saves the summary as a draft addressed to the leadership list, so nothing sends by itself. The reviewer opens the tracker for every line that says the brand dropped out, reads the stored answer, deletes anything the agent overstated, and then presses send.
6. The record. A Google Sheets row each week with the share of prompts naming the brand per assistant, so the trend outlives the inbox.

How Make bills scenarios and what the agent module costs in credits is in how to use Make.

The free version: a sheet, three assistants and two webmaster tools

Pick 20 prompts. Every Monday, ask each one in ChatGPT, Perplexity and Gemini from the same browser profile, and log one row per answer: named or not, where in the answer, which rivals appear, which sources are cited. Paste the answer text into the sheet as well; a screenshot cannot be searched three months later. That is the whole job, done by hand, and for a 20-prompt list it is enough.

Two free reports fill in what the sheet cannot see. Google counts traffic from AI Overviews and AI Mode in Search Console's Performance report, under the Web search type, as part of overall search traffic (AI features and your website). Bing Webmaster Tools, free to start (Bing Webmaster Tools), has an AI Performance report that shows which of your pages are cited in Microsoft Copilot, AI summaries in Bing and some partner integrations, with the grounding queries behind those citations (AI Performance).

The paid trackers add daily collection instead of weekly, more assistants, the stored answers, sentiment and citation counts, and a server an agent can read. The agent adds the Monday page.

What it costs

One gap to settle in a trial: Nightwatch's MCP page puts the server on Professional and above, while its pricing table lists the SEO MCP server on every plan. Budget for Professional at €199 a month until the trial shows otherwise. On Make's Free plan the agent runs on Make's AI provider, and that token use is billed in the same credits (MCP Client), so read the credits on the first run before trusting the 1,000.

Prompt tracking is sold by volume, collected answers at Nightwatch and prompts a day at SE Ranking, and either vendor does better as the list grows past the questions a buyer would ask. A prompt nobody would type is a cost with no reader. Keep the list short enough that the reviewer could read every answer if the agent broke.

How to know it worked

The measure is the share of tracked prompts whose answer names you, per assistant, week on week, set against the four-week baseline. Read it per assistant, never pooled: each one is a separate audience, and a pooled number hides which of them moved.

Take a holdout so the trend means something. Split the prompts by topic. Work on content for half the topics, the pages the sources keep citing, and leave the other half alone for a quarter. If the share rises on the worked topics and stays flat on the rest, the work moved it. Bing's own documentation warns that its citation trends are observational and cannot be attributed to any single update or content change (AI Performance); a holdout is how you get past that.

Then look for the money end. Search Console clicks on the pages behind the worked topics, and the self-reported source field on your forms, the one that asks where a buyer first heard of you, show whether being named sends anyone.

FAQ

Why not ask ChatGPT myself each week?

That is the free version, and for 20 prompts it works. A tracker adds daily answers across more assistants, keeps them for later reading, and exposes them to an agent, which matters once the list passes what one person reads on a Monday.

Nightwatch or SE Ranking for the tracking?

Nightwatch if rankings and AI answers are the whole job and seats matter, since unlimited seats start on Professional (Nightwatch pricing). SE Ranking if you also want keyword and backlink research in the same login. Nightwatch vs SE Ranking weighs the two in full.

Can the agent change anything in the tracker?

Not with a Nightwatch key, which is read-only unless you opt into write scopes. An SE Ranking key has full account access, so the protection is choosing only reading tools in Make's MCP Client and keeping the key out of any other scenario.