The short answer
Yes. Start with Google Search Console and Bing Webmaster Tools for the AI-search activity they cover. Add a small set of real buying questions, repeat them under recorded conditions, and save the answers in a spreadsheet. This provides a bounded baseline, not a census of every AI conversation. Free tools still take time to use well.
The first useful AI-visibility report can fit in a spreadsheet. The hard part is deciding which questions deserve rows.
You do not need to buy a subscription to discover that your own brand name was included in every prompt. You do need a measurement you can explain before automating it.
Start with data you do not have to simulate
Google announced dedicated Search Generative AI performance reports in June 2026, with worldwide rollout noted on 31 August. They show impressions and page-level visibility within Google’s generative search features, with country and time breakdowns. They do not measure every conversation across other assistants.
Bing’s AI Performance update adds intent, topic, citation-share and period-comparison views for supported experiences. Its citation share relates to citations for a grounding query; it is not brand recommendation share or traffic share.
Use the reports available for your verified property and record their scope beside each number. Separately, inspect referral visits and useful on-site outcomes in your analytics. A citation without a click will not become a referral visit. A blank chart is a reason to check coverage and dates, not a verdict on your brand.
Choose a small panel you can maintain
Pick five buying questions from sales conversations, support requests or relevant search queries. Remove private details. Keep your brand out of the discovery questions. Make the market and audience specific only where they affect the decision.
For example, a studio might ask which tools help a small team manage client approvals. “Which platform has our exact list of differentiating features?” is a different test. Our prompt-selection guide includes a matrix for keeping those decisions straight.
As a small illustrative exercise, five questions across two assistants, repeated three times, gives 30 answers to inspect. Those counts are a manageable starting design, not a statistical minimum or a claim of representativeness. Start new conversations and record the interface, date, search mode and relevant account settings. Keep API and consumer-app runs separate.
Use one row per answer
| Column | What belongs here |
|---|---|
| Question ID + exact wording | The fixed buying question; not a keyword you change each week. |
| Provider + interface + date + repeat | Enough detail to distinguish comparable runs. |
| Brand named / recommended / first | Separate yes/no fields; preserve the counting rule. |
| Other brands and cited URLs | Who appeared instead, and which source pages were referenced. |
| Full answer + claim to check | The evidence and any consequential uncertainty. |
| Action + owner + check date | One useful next step, if the evidence supports it. |
If your brand appears in 6 of 15 valid answers from one provider, its mention rate in that sample is 40%. Do not rename that “40% of the market”. Count repeated use of the same name within an answer once. Keep provider rows separate before considering a pooled view.
To calculate share of brand appearances, you need the competitor counts too. The share-of-voice calculator shows why that denominator can change your result even when your own count stays the same.
Repeat it when there is a decision to make
Choose a cadence you can maintain and a reason for reviewing the result. A monthly content review may need a monthly benchmark. A fast-moving launch may justify closer checks. Repeating the exercise every morning without changing any decision mostly creates more spreadsheet.
Preserve the original question set for comparisons. Add newly discovered questions to a separate exploration tab, and establish a new baseline if they later join the main panel. Log website changes and known provider changes alongside the dates.
Read recurring patterns: the same competitor solving the same buyer need, an inaccurate claim repeated across runs, or a source page worth inspecting. A one-off mention is still worth saving, but it is a poor reason to rewrite the website.
Pay when the work becomes the bottleneck
Automation earns its place when collecting, checking and presenting the evidence takes more effort than acting on it. Before paying, ask whether you can inspect the prompts, original answers, source links, repeat handling and what happens to failed requests.
We sell a one-off report, so this is not disinterested purchasing advice. If your spreadsheet already answers the decision in front of you, keep it. If you want a broader, repeatable collection with a reviewed action plan, compare the sample against the evidence you currently have. The useful upgrade is a better decision, not simply more rows.
Sources & context
This guide responds to a question raised in Do you track AI visibility in any serious way, or is everyone just guessing? on Reddit. The discussion informed the topic; it is not evidence of the effectiveness of any product mentioned there. Platform documentation is linked alongside the claims it supports.
Examples and figures are illustrative unless explicitly stated otherwise. Written and published by Share of Voice, a Superstellar LLC product. We sell AI visibility reports; our commercial perspective is worth keeping in mind.
Spotted something we should correct? Let us know.