How to Start AI Visibility Monitoring
The first 30 days of AI visibility monitoring for B2B SaaS teams: question set, baseline scan, ownership, monthly cadence and first decisions.
Short answer: You do not need a big tool budget to start AI visibility monitoring. The first month comes down to four jobs: write down the questions your buyers ask, run a first scan under fixed conditions, give the work to one owner and set a monthly cadence. Only when those four are done do you make the first decisions.
We explained how monitoring works, the measure, interpret and act steps, in our AI visibility monitoring guide. This post is the practical side. It covers what a B2B SaaS marketing team does in the first 30 days, and in what order.
Three decisions before you start
Answer three questions before you open the calendar. If they stay open, the first scan is wasted.
- Which product, which category? If you have several products, start with the category where most of your sales happen. If you monitor everything at once, the question set swells and the first month's results get hard to read.
- Which market, which language? A buyer asking in Turkish in Turkey and a buyer asking in English abroad may see different answers. If you sell in two markets, keep two question sets.
- Who owns it? Monitoring without an owner is the kind of work that is forgotten by the second month. We cover this decision separately below.
Week 1: build the question set
The question set is the foundation of the measurement. You will not change it for the rest of the month, so give it the first week.
A B2B SaaS buyer looking for software usually asks five kinds of questions:
- Category: "What is the best CRM for small sales teams?"
- Problem: "How do I spot customer churn early?"
- Alternatives and comparisons: "What can I use instead of X?", "X or Y?"
- Integration and compliance: "Project management tool that works with Slack", "GDPR-compliant e-signature software"
- Pricing and selection: "How is pricing set in this category?"
Do not invent the questions at your desk. The best sources are sales call notes, the "how did you hear about us" answers in your demo form and support tickets. Write in the buyer's own words. If your buyer does not say "CRM software" but "a customer tracking tool for my sales team", write the question that way.
Keep branded and unbranded questions apart. A question that contains your brand name brings up the brand by itself. Unbranded questions are the ones that measure visibility. 20-30 questions are enough to start. Next to them, list the 3-5 competitors most likely to appear with you for those questions. Give each question a row in a table: question, language, question type, branded or unbranded.
Check technical access in the same week. If an engine cannot read your page, the measurement only tells you "we are not there", not why. In its documentation on its crawlers, OpenAI writes that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Google's AI features page says that to be a supporting link in AI Overviews and AI Mode, a page must be indexed and eligible to be shown in Search with a snippet. According to Google, there are no additional requirements or special optimizations for this. So the first checklist is short: robots.txt, index status and whether key pages can be read without running JavaScript.
Point the tool at your own site first. When we did, we found a cosmetic effect on our homepage that was delaying performance. The details are in our guide.
Week 2: run the first scan
The first scan is your baseline. You will read every later measurement against it, so write the conditions down:
- Engine set. State at the top of the report which engines you measured. If an engine did not run in the scan, enter "not measured" instead of a zero.
- Repetition. Ask each question to the same engine more than once. AI answers are generated fresh every time. A single answer is a sample, not a measurement. We explain why in our post on answer variance.
- Clean session. Check while logged out, in a session without chat history, and note the location. The answer seen by someone who searches for your brand every day may not be the answer your buyer sees.
- Three numbers per engine. For each engine, write the number of responses, the mention rate and the margin of error separately. As our methodology page states, in a scan of 200–300 answers the margin of error is roughly ±4–5 points per engine.
- Source list. Record which pages each answer cites. Your own page, a competitor's page and a list article all belong on this list.
Search Console does not replace this work. Google writes on the same page that AI Overviews and AI Mode traffic is reported in the Performance report, within the "Web" search type. There is no separate AI line. To see whether you are mentioned inside the answers, you have to ask the engines the questions.
Who should own it
In B2B SaaS teams this work usually lands with product marketing or the content lead. Whoever it is, it should be one person. When the work is split across several people, nobody keeps track of the scan date.
Write down the roles at the start:
- Owner: Runs the scan or runs it with an agency, writes the report and keeps the change log.
- Content team: Does the page and content work that comes out of the report.
- Sales and customer success: Pass new buyer questions to the owner. These teams keep the question set alive.
- Leadership: Sees a one-page summary once a month. It reads the status per engine and the decisions, not the raw tables.
Week 3: set the cadence
Cadence matters more than frequency. We run a full scan once a month. In between, only check the related questions for pages you changed.
Open the change log this week. Every piece of content you publish, every page you fix and every press release you send becomes a row: date, URL, what changed. Two months later, only this log can answer the question "why did visibility go up?".
Do not change the question set within the month. If you want to add a question, mark it as a separate series. If you mix it into the same average as the old questions, you cannot compare two measurements.
Week 4: make the first decisions
The table from the first scan is in front of you. Check which of these four cases you are in:
- You are not mentioned at all for unbranded questions. That is a content gap. Check whether you have a page that answers the buyer's question in its first paragraph. If not, that page is your first content job.
- A competitor is mentioned and the engine cites a list or comparison page. Engines often summarize a list someone else wrote. In our measurement of 29 October 2026, 375 of 476 answers cited at least one agency list page. In this case the job is to be listed in those sources with correct information. The details are in our 476-answer measurement post.
- You are mentioned, but the information is wrong or outdated. The engine may have found an old page of yours. Find and fix the pages that still carry an old price, a feature you removed or an old product name.
- The difference is smaller than the margin of error. Do not decide. Wait for the next scan.
There are also three mistakes to avoid in the first month. Do not raise an alarm over a single screenshot. Do not change the question set every week. Do not split the budget by looking only at the composite score.
Next step
By the end of the first month you have a question set, a baseline scan, an owner and a change log. From the second month on, the same workflow repeats every month.
If you want to set up the first scan and the first decisions with us, see our GEO service for B2B SaaS. Packages and scope are on our pricing page.
