AI visibility is difficult to measure because there is no single report that shows every time a system names your business, cites your website, or describes your services. A mention without a click may leave no trace in your analytics. Referral traffic captures only some of the people who do click through.
The measurement gap is real. It is also manageable. With a consistent prompt set, a simple tracking table, and a clear understanding of what current tools can and cannot tell you, you can build a useful baseline and review how it changes.
This is the natural companion to our guide on why ChatGPT may not know your business. That article explores the visibility problem. This one focuses on how to tell whether it is changing.
Four metrics worth tracking
The terms can sound interchangeable, but each describes a different signal.
Mentions: Were you named at all?
A system may identify your business in an answer without linking to your site. Count a mention when your business is named, whether or not a source is provided.
Citations: Was your site linked as a source?
Record a citation when the answer links to a page on your website. A mention and a citation are not the same: you could be named without being linked, or your page could be cited without your brand being discussed prominently.
Share of voice: How often were you named compared with competitors?
Use the same prompts to track your business and the competitors a buyer might consider. Calculate your share from the business names mentioned in those answers. Keep the competitor set and calculation consistent from month to month.
Accuracy: Was the description correct?
Check facts such as your services, location, audience, offers, and credentials. Frequency alone is not a good measure of visibility quality. A wrong description can be worse than no description: it may send a prospect away with the wrong understanding of your business.
Build a repeatable prompt set
Start with 15–20 questions a buyer might ask while researching a provider. Include prompts about recommendations, comparisons, your category, the problem you solve, and your brand.
For a 15-prompt set, you could create three prompts in each group:
- Recommendation: “Which [category] businesses serve [location]?” “Who can help with [specific need]?” “What providers should I consider for [buyer goal]?”
- Comparison: “[Your business] vs. [competitor].” “Compare [provider A] and [provider B] for [need].” “What should I consider when choosing between [options]?”
- Category: “What does a [category] provider do?” “What should I look for in a [category] company?” “What services are usually included in [category]?”
- Problem: “How can a business address [problem]?” “Who helps companies improve [specific outcome]?” “What should I do if [common challenge]?”
- Brand: “What does [your business] do?” “Where does [your business] operate?” “What services does [your business] offer?”
Shape the wording around real buyer language and your market. Avoid building the set entirely around your brand name; you also want to understand whether you appear when people ask category and problem questions.
Once you have selected your prompts, keep them fixed. If the wording changes every month, it becomes harder to tell whether visibility changed or the test did.
Run each prompt in a fresh conversation on each platform you are testing. Avoid follow-up questions, which can add context and influence later answers. Log the platform and date, and note meaningful changes such as a new model version or a change in your test location.
Record what happened
A spreadsheet is enough to begin. Use one row for each prompt and platform combination. Save the original answer or a screenshot alongside the log so you have evidence to compare later.
| What to record | How to capture it | Why it matters |
|---|---|---|
| Prompt and platform | The fixed prompt text and the engine tested | Keeps the test consistent month to month |
| Mention or citation | Whether the business was named, and whether a page was linked | Separates awareness from linked authority |
| Competitor named instead | The competing business, if any | Shows who you are being measured against |
| Accuracy notes | Any factual errors, outdated details, or missing context | A wrong description can cost you a prospect |
| Raw output | A saved answer or screenshot, filed with the date | Gives you evidence to compare at the next review |
To make this easier, we have prepared a one-page AI Citation Tracking Log (PDF) you can download, print, or copy into your own spreadsheet. It includes the four metrics, five starter prompts, and a ready-to-fill tracking table.
Add a brief note when the answer contains a factual error, cites an outdated page, or names a competitor in your place. Keep the raw outputs in a dated folder or document. A short record of what the system actually returned makes the next review more reliable.
Read current reporting honestly
Google now provides a dedicated Generative AI performance report in Search Console. Google says it became available to all websites worldwide on August 31, 2026. The report covers impressions from AI Overviews and AI Mode in Google Search, with views by page, country, device, and date.
Treat it as a signal of presence, not a complete citation report. It shows when links to your site were shown in Google’s AI features. It does not provide clean performance attribution for each citation or tell you how often your brand was mentioned without a link. Google’s announcement of the reporting surface describes impressions as the central measure and notes that additional metrics may develop over time.
Referral visits from AI assistants can provide another partial signal. If someone clicks through from an assistant to your site, the visit may appear in your analytics as a referral. But this captures only people who click, and referral information can be missing or classified differently. A visit tells you that traffic arrived; it does not tell you how many times your business was discussed without a click.
For now, Search Console, analytics, and prompt testing each show a different part of the picture. None provides clean per-citation attribution across AI search.
Why measurement is becoming a discipline
The standards around content and its use in AI answers are evolving. Google’s current people-first content guidance warns against fabricated author profiles, such as made-up names, false credentials, or AI-generated headshots presented as real experts. Accurate authorship and dependable information matter to readers, not just measurement systems.
Google has also been testing an AI Contribution program with a limited group of publishers. As reported by Digiday, the pilot explores payments when publisher content meaningfully contributes to AI-generated answers. It is not a general citation-reporting tool, and its terms and availability are limited.
These developments do not replace the need for your own baseline. They reinforce why visibility should be observed consistently and described carefully.
Report trends, not snapshots
A single answer is a snapshot. Models can vary their responses, so one query on one day is not a dependable trend.
A set of ten priority prompts checked monthly for six months can offer a more useful signal. It will not prove why visibility changed or show every exposure. It can help you see whether your presence is becoming more consistent, which competitors appear, and whether your descriptions are becoming more accurate.
For example:
“We appeared in 40% of our tracked prompts this month, up from 15% at baseline.”
That statement defines the sample and reports a measured change. By contrast, “We now rank for AI search” implies a stable position and a ranking system that the test does not establish.
When reporting to leadership, state the scope and confidence plainly:
- How many prompts and platforms were tested.
- Whether the prompts and test conditions stayed consistent.
- What changed since the previous review.
- Where answers remain variable or the data is incomplete.
If you want to make that review part of a dependable operating rhythm, our guide to marketing operations strategy covers how to create a useful review cadence. AI visibility also belongs within one system, three surfaces, alongside SEO, answer visibility, and broader marketing infrastructure, not as a standalone replacement for them.
Turn a finding into a next step
When an answer names your business but describes it incorrectly, first check whether the same facts are clear and consistent across your website and other trusted profiles. If you are named but not cited, review whether the relevant page directly answers the prompt and provides clear supporting information.
Keep the correction focused. Our Generative Engine Optimization guide covers the implementation work in more detail. The wider aim is alignment: visibility, content, website structure, and follow-up should support the same customer journey. That is the principle behind a connected marketing system.
Measure the current picture. Then improve it.
A prompt set and a tracking table can give you a practical starting point. The next step is understanding where your business appears across platforms, how it is described, which competitors are present, and what to address first.
The Holly Acre Media AI Visibility Audit tests your brand across four AI search platforms, benchmarks it against three named competitors, and delivers a scored report with visibility gaps and a recommended fix order. It is a structured way to turn scattered observations into a clearer baseline.
Holly Acre Media connects search visibility, strategy, websites, content, CRM, automation, and ongoing improvement into one coordinated system. Explore our services or send us a message to discuss a practical next step.
Less noise. More direction. Clearer visibility.
