Key Takeaways
- AI-search visibility is not one reliable number. Measure citations, brand mentions, referral visits, page visibility and business outcomes separately.
- Use first-party platform data where it exists, then add a repeatable prompt panel for the gaps.
- Record the exact question, platform, date, market and result. A screenshot without that context is difficult to compare later.
- Referral traffic shows visits, not every mention. AI answers can influence a decision without producing an immediate click.
- Connect visibility work to qualified enquiries, assisted conversions and sales questions so the report does not become a vanity score.
Measuring visibility across Google AI Overviews, ChatGPT, Perplexity and Copilot is possible, but it is not as tidy as checking one ranking report.
Each platform has different interfaces, data access and answer behaviour. Results can change with the wording, location, account and date. A brand might be cited in one answer, mentioned without a link in another and absent when the question is slightly broader.
The solution is a small measurement system with three layers: platform reporting, website analytics and a controlled prompt panel. Add commercial outcomes and you have something useful enough to guide content decisions.
Define Visibility Before Choosing a Tool
Start with a written definition. For this report, AI-search visibility can mean any of the following:
- a linked citation to your domain;
- the business or product name in the generated answer;
- a fact from your page used in the answer;
- the domain in a supporting-source panel;
- a referral visit from the platform;
- an enquiry in which the person says an AI assistant helped them find you.
Do not merge those events without labelling them. A visible citation is not the same as a click, and a visit is not the same as a qualified enquiry.
Set the audience and market as well. A Reading agency may want questions from UK small-business owners, not a worldwide set of generic technology prompts. For an ecommerce client, the panel may need separate informational, comparison and product questions.
Our answer engine optimisation guide explains why the work starts with real user questions rather than a collection of fashionable phrases.
Build a Core Prompt Panel
A prompt panel is a fixed group of questions checked repeatedly. It gives you a comparable sample when the platform does not provide a complete reporting export.
Create four groups.
Problem Questions
These describe an issue before the buyer knows the solution:
- Why did my WordPress traffic fall after a redesign?
- How can a UK business reduce repetitive support enquiries?
- Why is my Shopify category page not ranking?
Category Questions
These ask about the type of service or product:
- What should an SEO audit include?
- Which AI customer-service tools suit a small UK team?
- What is involved in migrating WordPress to Next.js?
Comparison Questions
These show a decision is forming:
- Is Next.js faster than WordPress?
- HubSpot Content Hub or WordPress for a growing service business?
- Should we use a custom AI integration or an existing tool?
Recommendation Questions
These can expose local and commercial visibility:
- Which agencies provide WordPress rebuilds near Reading?
- Who can help a UK small business integrate AI with HubSpot?
- What should I look for in a local SEO company?
Begin with 20 to 30 questions. Keep a stable core so the trend means something. Add a smaller rotating group from recent Search Console queries, sales calls and customer-support conversations.
Record the Test Conditions
For every check, store:
- exact prompt;
- platform and product surface;
- model or mode if displayed;
- signed-in or signed-out state;
- country and language;
- date and time;
- brand mentioned: yes or no;
- domain cited: yes or no;
- cited URL;
- position or prominence, described consistently;
- named competitors;
- screenshot or saved answer;
- short note on accuracy.
Do not ask the same question ten times and select the answer you like. Decide the method first. For example, run each prompt once per platform each month from the same UK setup. If you repeat prompts to study variability, record every result.
Be careful with personalisation. An account that has already visited your website may not reflect a new customer's experience. A private or signed-out session can reduce some history, but it does not make every platform test identical.
Use Google's First-Party Reporting
Google is rolling out a dedicated generative AI performance report in Search Console to a subset of website owners. The Google Search Console help page says it reports impressions for AI Overviews and AI Mode with page, country, device and date dimensions. The report may be absent when a property has not received enough impressions, and Google does not describe clicks in this dedicated report.
Use it to answer:
- Is AI-feature visibility rising or falling?
- Which pages appear?
- Which countries and devices produce impressions?
- Are impressions concentrated on informational pages or commercial pages?
- Did a content change or migration affect the impression pattern?
Keep the ordinary Search performance report beside it. Compare total clicks, impressions, queries and pages without assuming that a movement in one report caused a movement in the other.
Google's AI features guidance says the same SEO foundations apply to AI experiences. This means URL Inspection, crawl status, canonical selection, internal links, content quality and snippet eligibility remain useful diagnostic evidence.
Track Referral Visits Correctly
Website analytics can show people who click a link from an AI platform. Create a channel or exploration that identifies known referring domains and inspect the landing page, engagement and conversions.
OpenAI's publisher and developer FAQ states that ChatGPT adds a referral parameter to search links, which can help analytics identify that traffic. Verify the actual data in your own property and keep the classification rule documented.
For each AI referral source, review:
- sessions and users;
- landing pages;
- engaged sessions;
- enquiry or purchase events;
- assisted conversions;
- new versus returning visitors;
- country and device;
- consent and tracking limitations.
Do not expect large click volumes simply because a citation appears. A generated answer may satisfy the question without a visit. That is why zero-click SEO needs brand and commercial measures beyond sessions.
Ask New Enquiries How They Found You
Analytics attribution is incomplete. Add one plain optional field to an enquiry form or sales process:
How did you first hear about us?
Include choices such as Google, ChatGPT or another AI assistant, recommendation, social media and other. Keep a free-text option. Train whoever handles calls to ask naturally when it helps.
Do not force a buyer to identify the exact path. They may have seen an AI answer, searched the brand, read reviews and then returned directly. Store “self-reported source” separately from last-click analytics.
In a CRM, add:
- first known source;
- self-reported source;
- landing page;
- service of interest;
- qualified or unqualified;
- eventual value or stage.
This lets you see whether a small amount of AI-search traffic brings stronger enquiries than a larger amount of general information traffic.
Create a Simple AI Visibility Scorecard
Avoid false precision. A small business can use a transparent scorecard rather than a mysterious percentage.
For each prompt, award:
- 2 points for a direct linked citation;
- 1 point for an accurate unlinked brand mention;
- 0 points for no presence;
- a separate accuracy flag;
- a separate commercial-relevance rating.
Then report:
- citation coverage: prompts with a citation divided by prompts tested;
- mention coverage: prompts with a citation or mention divided by prompts tested;
- accuracy rate: visible answers that describe the business accurately;
- commercial coverage: buyer-intent prompts with a citation or mention;
- unique cited pages;
- competitor citation share within the same sample.
The weights are a management choice, not an industry truth. Keep them stable and publish the method beside the result.
A score of 40% means “the brand appeared in 40% of this recorded panel under these conditions”, not “the brand owns 40% of AI search”.
Evaluate Third-Party Tools Carefully
AI visibility platforms can automate prompt checks, competitor comparisons and citation exports. Before buying one, ask:
- Which platforms and markets are tested?
- Can you supply your own prompts?
- How frequently are they run?
- Are logged-in, location and model settings documented?
- Are full answers and source URLs retained?
- Can results be exported?
- How are retries and answer variability handled?
- Does the tool separate citation, mention and sentiment?
- Can you inspect the evidence behind an aggregate score?
Run a short trial using prompts you already know. Compare the tool's output with manual checks. It may be useful even when results differ, provided the method is consistent and the limitation is understood.
Our analytics reporting service can combine this sample with Search Console, website behaviour and lead data. The measurement should support a decision, not simply create another dashboard.
Turn the Report Into Content Decisions
The most useful field is often the cited URL.
If AI systems repeatedly cite an old blog post rather than the current service page, improve the relationship between them. If competitors appear for cost questions, inspect whether they publish clearer assumptions. If your brand is mentioned inaccurately, correct first-party facts and important external profiles.
Classify gaps:
- access gap: crawler blocked or content unavailable;
- index gap: page not indexed or canonicalised elsewhere;
- answer gap: page does not answer the prompt;
- evidence gap: claim exists only on your website;
- commercial gap: informational visibility does not lead to a relevant service;
- measurement gap: no reliable way to connect the result to behaviour.
Choose one or two gaps per month. A content marketing programme should improve clusters and service pages, not flood the site with near-duplicate AI articles.
A Monthly Reporting Routine
1. Freeze the Core Panel
Keep the main questions and conditions stable. Add the month's new questions in a separate group.
2. Export First-Party Data
Collect Google AI reporting where available, normal Search Console data, website referral visits, landing pages and conversions.
3. Run the Prompt Checks
Test the four platforms consistently. Save the answers and source URLs. Mark obvious inaccuracies.
4. Review Leads
Add self-reported sources and qualified outcomes. Speak to sales or whoever receives enquiries.
5. Decide Three Actions
Choose a technical fix, content improvement or external-profile correction supported by the evidence. Assign an owner and date.
6. Keep the Raw Record
Store the exports, prompt sheet and methodology. A chart without its source record will be difficult to audit in six months.
For a wider search review, SEO services can connect AI visibility to organic search, service-page performance and technical health. If you want help building the first scorecard, contact MattDarm.
Frequently Asked Questions
What is AI search visibility?
AI search visibility is the extent to which a brand, website or source appears in answers produced by AI-assisted search experiences for a defined set of relevant questions. It can include a cited link, an unlinked mention, a quoted fact or visibility in a supporting-source panel. The metric only makes sense when the question set, market and testing method are recorded.
Can Google Search Console show AI Overview visibility?
Google is rolling out a generative AI performance report to a subset of website owners. It reports impressions for AI Overviews and AI Mode with page, country, device and date dimensions. Google does not describe clicks in this dedicated report, and a property may not receive it if it has too few impressions.
How can I track traffic from ChatGPT?
Use web analytics to review referral sessions from ChatGPT and tag any links you control with suitable campaign parameters. OpenAI documents a referral parameter used on ChatGPT search links. Referral traffic captures visits, not every brand mention or citation, so combine it with a repeatable manual prompt panel.
Are third-party AI visibility scores accurate?
They are samples, not a census of everything users see. A score can be useful for trend and competitor comparisons when the tested prompts, locations, models and dates are clear. Do not present it as exact market share, and retain screenshots or source records for important findings.
How often should a small business measure AI-search visibility?
A monthly review is usually enough for a small business, with a deeper quarterly analysis. Run the same core prompt panel under consistent conditions, then add a small set of new customer questions. Daily checks can create noise unless a migration, campaign or technical incident needs closer monitoring.
Measure the Decision Journey
AI-search reporting is still developing. That is a reason to document the method, not to wait for a perfect tool.
Use official data where it exists. Sample the unanswered areas consistently. Then connect citations and mentions to visits, qualified enquiries and customer language. The result will not be one magical number, but it will show which pages and facts deserve the next piece of work.
