How to build an AI visibility baseline for a client

Before you change anything for a client, measure where they stand. A baseline turns “we think things improved” into a comparison you can show. It also makes problems visible early, such as a competitor that dominates a topic or a crawler that's blocked.

The steps below work for most clients. The key is to make each choice deliberately, write it down, and keep it fixed long enough to compare.

1. Define the brand and its competitors

List the client's name and every alias that should count as a mention: short names, product names, common misspellings. Note any names that should not count, such as a generic word or an unrelated company with a similar name. Then pick three to five competitors the client actually loses business to, and list their aliases too.

2. Write the question set

Write the questions a buyer might ask an assistant before choosing a provider. Aim for a mix rather than a list of variations on one query. Twenty to thirty questions is a manageable starting point for most clients.

• Category questions: “What are the best accounting firms for small restaurants?”
• Problem questions: “How do I fix a furnace that keeps shutting off?”
• Comparison questions: “Is a physiotherapist or a chiropractor better for back pain?”
• Local questions, where location matters: “Who offers same-day dental appointments in Riverton?”
• A few branded questions, to see how the client is described: “Is Northwind Dental good for families?”

Keep branded questions to a small share. Most of the value is in seeing whether the client appears when buyers don't already know its name.

3. Choose engines and record how answers are collected

Pick the engines that matter for the client's audience. For each one, note how answers are collected, for example through an API with web search enabled or from a search results page. Different collection methods can produce different answers, so the method is part of the baseline.

4. Sample more than once

AI answers vary between runs, so one answer per question isn't enough. Ask each question several times per engine in the same period, and calculate rates across all the answers.

5. Record the baseline

For the baseline period, record the following for the client and each competitor, overall and per engine:

• Mention rate and recommendation rate, with the number of answers
• Citation coverage: the share of answers citing at least one owned page
• The most-cited domains, grouped by type
• The questions where the client is missing but competitors appear
• Collection dates, engines and collection method

Keep the answers themselves. When a number changes later, the first thing anyone will ask is “what did the answer say?”

6. Pair it with a search baseline

Record the search side at the same time: positions for tracked keywords, Search Console clicks and impressions, and a technical audit including AI crawler access. If the client later gains mentions and also climbs in search, you'll want both starting points.

7. Freeze the set, and annotate changes

Leave the question set, engines and competitor list unchanged for several weeks so trends compare like with like. When you do need to change them, note the date and treat the new set as a new baseline rather than mixing old and new results.

Baseline checklist

1. Brand and competitor aliases written down, including exclusions
2. A mixed set of 20–30 buyer questions
3. Engines and collection methods recorded
4. Several samples per question per engine
5. Rates recorded with their answer counts, overall and per engine
6. Answers and cited sources saved
7. Search positions, Search Console data and a technical audit captured
8. Set frozen, with a note of any future changes