Are we appearing in AI answers, and is it bringing us business?
Owners ask this in every planning meeting now. The first answer is usually a screenshot. Someone types a question into an AI tool, sees the company named, and shares the image. It feels like proof. It is one answer, at one moment, for one person. Ask again from another city, with slightly different words, and it can change.
A screenshot cannot tell you how often you appear. It cannot tell you whether anyone clicked, or whether anyone called. A monthly report can. This guide shows you how to build one.
Our posts on answer engine optimization and GEO for small businesses explain why AI answers matter. This one covers measurement.
Six terms to keep apart
Confusion starts when one number is used to answer a different question.
| Metric | What it shows | What it cannot prove |
|---|---|---|
| Impression | Your link was shown to a user in an AI feature. | That anyone read it, clicked it or remembered you. |
| Mention | An AI answer names your business in its text. | That the description is accurate, or that a visit followed. |
| Linked citation | Your page was listed as a source in an answer. | That anyone clicked, or how prominent the source was. |
| Referral session | A visit arrived from an AI platform. | That you are seeing every AI visit. Some carry no referrer. |
| Enquiry | A form, call, booking or message arrived. | That the person was a good fit. |
| Qualified lead | An enquiry that meets your own test: right service, right area, right budget. | That the lead will buy. |
What Google and Bing report today
Google. Search Console now has a Generative AI performance report (Search) under Performance. It covers AI Overviews and AI Mode. Its one metric is impressions, defined as how many times links to your site were shown to a user in a generative AI feature. According to Google’s help page, it does not show query data, click data or citation data. Google says the report reached all websites worldwide as of August 31, 2026, and the newest data is preliminary and may change within hours.
Bing. AI Performance in Bing Webmaster Tools launched in public preview on February 10, 2026. It reports total citations, average cited pages, grounding queries (the phrases the AI used when it retrieved cited content), page-level citation activity and visibility trends. Bing says the grounding queries are a sample. It also says the counts reflect how often you were cited, not how important or prominent the page was.
A June 16, 2026 update added four views, rolling out in preview. Intents and Topics group grounding queries. Compare overlays two time periods. Citation Share is the percentage of citations going to your site out of all citations shown for a query. Bing states it is not a ranking system and does not represent traffic share.
Two points follow from this.
First, the units differ. Google counts impressions. Bing counts citations. Record them in separate rows and never add them together.
Second, everything in these reports is observed visibility. Traffic and revenue are inferred until your own analytics and enquiry records confirm them. Rising impressions with falling clicks is not proof of AI citations either, so do not report it as one.
Google also states in its AI search guidance that optimizing for generative AI search is still SEO from its point of view, and that no third-party tool can see its internal ranking or AI systems. Treat SEO, AEO and GEO as overlapping work. Measurement is where they start to differ.
Find AI referrals in your analytics
Open the traffic acquisition report in your analytics tool and look at session source. Build a saved segment for the AI platforms you see sending visits, such as chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Hostnames change, so review the list every quarter.
Then connect those visits to business outcomes:
- Track form submissions, bookings and phone taps as key events.
- On a WooCommerce store, track completed purchases with the same segment.
- Add a “How did you find us?” field to every form, with “AI assistant” as an option. Ask the same question on the phone.
Expect gaps. Some AI visits arrive with no referrer and land in direct traffic. Clicks from Google’s AI features may be counted with ordinary Google traffic. Samples are small. Six sessions and one enquiry is a note, not a trend. Report counts instead of percentages until you have several months of data.
Run a repeatable manual sample
- Write 10 to 15 questions your customers really ask. Pull them from enquiry emails, call notes and sales chats.
- Freeze the wording. Do not rewrite questions between months.
- Each month, run the same questions on the same platforms.
- Log the date, platform, location context (city, signed in or out), a short summary of the response, whether your business was named, the cited URL, and whether the description of your business was accurate.
This is a sample of your customers’ questions, and it is no universal ranking. Repetition gives it value. Three months of the same twelve questions tells you more than any single answer.
A monthly dashboard template
Copy this into a spreadsheet. Leave the cells blank until you have real figures.
| Metric | Source | Current period | Prior period | Limitation | Next action |
|---|---|---|---|---|---|
| Google AI impressions | Search Console, Generative AI performance report | Impressions only. No clicks, queries or citations. | |||
| Bing total citations | Bing AI Performance | Citation counts only. No prominence or ranking. | |||
| Bing Citation Share, top three queries | Bing AI Performance | Not traffic share. Preview feature. | |||
| Sample questions naming the business (out of 12) | Manual log | Small sample. Answers vary. | |||
| Sample answers describing the business accurately | Manual log | Judged by a person. | |||
| AI referral sessions | Analytics | Missing referrers undercount. | |||
| Enquiries from AI referral sessions | Analytics and forms | Small numbers. | |||
| Enquiries naming an AI tool in “How did you find us?” | Form and call notes | Self-reported. | |||
| Qualified leads from the two rows above | Your own review | Needs your definition of qualified. |
Illustrative example: a Calgary furnace repair company
This example is fictional. The company, the figures and the decision are invented to show the method. They are not Paradigmmedia client data or measured results.
Picture Foothills Heating Co., a made-up furnace repair business in Calgary. The owner sees 14 AI referral sessions in a month and decides AI search is not worth any effort.
Then the company adds a “How did you find us?” field and asks the same question on every phone call. Suppose the next month’s log shows the same 14 sessions, with 3 form enquiries from those visits. It also shows 6 more enquiries where the customer said they had asked an AI assistant first, and those people arrived by phone or by typing the web address. The referral report alone would never have shown them.
The manual sample adds a second finding. One AI answer describes the company as offering overnight emergency repair, which it does not.
The decision changes. The owner no longer writes off AI search. The team fixes the service page that states hours and service area, and runs the same twelve questions next month to see whether the description changes. No gain is claimed here. Better tracking changed what the owner chose to do next.
Reading the patterns
- Citations up, leads flat. Open the cited page. Does it answer a buying question or only an information question? Is there a clear phone number, form or booking step? Citation Share measures sources, not visitors.
- Impressions up, referral sessions flat. Google’s report gives impressions only. Do not fill the gap with a guess about clicks.
- Referral sessions up, enquiries flat. Check the landing page on a phone. Check form speed and whether the next step is easy to find.
- Mentions without links. Note them. Check whether the description is right.
- Everything flat. With small samples, one quiet month means little. Decide after three months.
Three next actions
- Open Search Console and Bing Webmaster Tools. Record this month’s figures in the template.
- Write your 12 customer questions and log the first round this week.
- Add “How did you find us?” with an AI assistant option to every form and phone script.
Request an AI search visibility baseline audit
Paradigmmedia builds and maintains WordPress and WooCommerce sites and plans search strategy for businesses in Toronto, Calgary and across Canada. No audit can promise rankings or AI recommendations. A baseline shows where you stand today, sets up the monthly report, and tells you what to track next.
Book a free consultation to request your baseline audit. You can also read about our GEO, AEO and SEO services.
Frequently asked questions
Do I need a paid tool to track AI search visibility?
No. Search Console, Bing Webmaster Tools, your analytics and a spreadsheet cover the basics at no cost. Google says third-party tools cannot see its internal systems, so check what any vendor measures and how.
How often should I check?
Report monthly. Run your manual sample on the same day each month. Judge trends over three months or more, because single months move around.
Is AI search visibility different from SEO?
The work overlaps. Google says optimizing for generative AI search is still SEO. The measurement differs, because AI features report impressions or citations and often send few clicks you can trace.