To track your AI visibility every week, ask the same set of buyer questions to the same assistants on a fixed schedule, record whether each answer names your brand, at what rank and from which sources, and compare each week with the one before. You can do it by hand, with a browser extension that asks the questions for you, or with an automatic run that a server performs while nobody is at their desk. The method matters less than the rule behind it: same questions, same assistants, same rhythm.
This guide compares the three approaches, explains why answers from an API and from an app can differ, and lists the five numbers worth reading every Monday.
Key takeaways
- Weekly beats daily for most brands: AI answers move slowly, and a weekly series is easier to read than a noisy daily one.
- Keep the question set fixed. Changing the wording of a prompt starts a new series.
- Ask each question more than once per run. One answer can be a fluke; two give you a steadier number.
- An API answer and an app answer are close, not identical. Use both when you can: a steady weekly trend from the API, the exact picture from the app.
- Read the change, not the level: who passed you, which assistant dropped you, which sources moved.
Why a weekly rhythm
AI assistants don’t refresh their view of your market every hour. Their answers change when the pages they search change, when competitors publish, or when the model behind them is updated. A weekly reading catches those moves without drowning you in day-to-day noise, and it gives a team one moment to look at the numbers together.
A weekly rhythm also makes cause and effect easier to see. If you publish a comparison page on Tuesday and your visibility on that question rises the following week, the link is plausible. With checks every few months, too many things change in between.
Option 1: check by hand
Open ChatGPT, Claude, Gemini and Perplexity, paste each of your buyer questions, and note in a spreadsheet whether your brand is named, its position in the answer and the sources cited.
It works for a handful of questions and costs nothing. It breaks down quickly:
- Time. Twenty questions in four assistants is eighty answers. Asked twice, a hundred and sixty. That’s a morning every week.
- Consistency. People rephrase questions without noticing, skip one, or check on a different day. The series stops being comparable.
- Personalisation. Signed-in apps adapt to your history. Your own account may know your brand better than a buyer’s does.
Manual checks are a good way to choose your questions. They are a poor way to follow them.
Option 2: a browser extension
An extension can open the assistants in background tabs, ask your questions with the accounts already signed in to your browser, and read back each answer and its sources. Citegram’s Chrome extension does this across nine assistants, including Google AI Overviews and AI Mode, which have no public API.
The strength of this approach is fidelity: you measure the answer the app shows, with its web search and its cited pages. The limit is that someone has to start the run, with Chrome open. That suits one person checking before a meeting. It’s harder to keep up every single week across a team, and the runs count toward your own assistant accounts.
Option 3: automatic weekly runs
The third approach takes people out of the loop. A server asks your questions to the assistants through their official APIs, with web search turned on, at the day and hour you choose, and stores the answers. Nothing to install, no browser open, no account of yours involved.
This is what Citegram’s Autopilot does on the Scale plan:
- Every week, at the slot you pick (weekday, hour and time zone), it asks your 20 tracked prompts to ChatGPT, Claude, Gemini or Perplexity. Two assistants are included; you can add a third or a fourth.
- Each prompt is asked twice per assistant, so a single unusual answer doesn’t swing the week.
- The day after the run, every member of the workspace gets a weekly email: visibility, share of voice and average rank against the week before, the brands named more often than you, the ones that passed you, and the next three actions.
- Citegram also writes 8 blog articles a month on the questions where the assistants leave you out, so the gaps the report shows turn into pages you can publish.
The trade-off is the one every API-based tool shares, explained below.
API answers vs app answers
An assistant’s app and its API run on related models, but the app adds layers: its own web search, formatting, memory of the conversation, and personalisation for the signed-in user. The API answer, even with search enabled, comes without some of those layers. So the two answers to the same question are usually close in substance and in the brands they name, and can still differ in order, wording or cited pages.
Two habits keep this from misleading you:
- Compare like with like. Read the weekly trend from the same method every week. A rise from 30% to 42% on Autopilot means something; comparing an API week with an app week doesn’t.
- Check the apps when it matters. Before a decision, run the same questions in the real apps with the extension. On Scale you have both: the automatic weekly series and the exact app picture on demand.
The five numbers to read every week
Whatever method you use, these are the numbers that tell you something changed.
| Number | What it tells you | What to do when it moves |
|---|---|---|
| Visibility | The share of answers that name your brand | A drop on one assistant: check which questions you lost there |
| Share of voice | Your mentions against all brand mentions | It falls while visibility holds: competitors are named more often next to you |
| Average rank | Where you appear when you are named | Rising rank number: you’re still listed, but lower in the answer |
| Who passed you | Rivals behind you last week and ahead now | Read the answers they win and the sources those answers cite |
| Sources | The pages assistants rely on | A new source names competitors and not you: ask to be listed, or publish the page it should cite |
Read them in that order: visibility for the headline, share of voice and rank for the detail, then the rivals and sources for the explanation.
A weekly routine that holds
- Monday morning: read the weekly report. Note the one change that matters most.
- Monday, 15 minutes: open the questions behind that change. Read two or three answers and the sources they cite.
- During the week: act on one thing. Publish or update a page, ask a cited site to list you, or answer the thread that keeps coming up.
- Next Monday: check whether that question moved.
One change a week, followed for a month, teaches you more than a quarterly audit of everything.
Getting started
If you only want to see where you stand today, the free plan tracks one question in one assistant. When you want a series, Pro tracks 20 questions in all nine assistants whenever you run the extension. For a team that wants the reading to happen on its own, Scale adds Autopilot, the weekly email and the monthly articles: see how Autopilot works.
