Measure AI search at three levels: what the answers say about your business, which visits can be attributed to those services, and what those visitors do next. Keep mentions, citations, recommendations, inquiries, and customers separate. A useful report shows the evidence behind each number and makes the next marketing decision easier.
A report can show encouraging visibility while leaving the owner with a practical question: what should we do next? The answer depends on what was measured. A captured recommendation, a website session, and a sales conversation tell you different things about the work.
Start by agreeing on the business decision the report needs to support. You may be deciding which service deserves a stronger page, whether to expand a content series, or why inquiries are not moving forward. Choose measurements that help with that decision and retain enough detail to check them.
Give each result a clear definition
Use these working definitions in your own reporting. When a platform supplies a metric, retain its exact label and counting method alongside your interpretation.
| Result | Evidence to keep |
|---|---|
| Mention | The saved answer names the business. Record what it says. |
| Citation | The answer identifies or links a source. Keep the cited URL and context. |
| Recommendation | The answer actually proposes the business as an option for the stated need. |
| Attributable visit | Your analytics records the visit with an identifiable source under its reporting method. |
| Qualified inquiry | Your team confirms the inquiry meets the service, geography, or other criteria you agreed to use. |
A citation to an educational article does not necessarily recommend its publisher for a service. A recommendation in one saved answer does not establish how often other people receive that recommendation. Write the finding at the level the evidence supports.
Keep inquiries and customers separate too. A submitted form is an action to inspect. Decide whether it represents a real request, an appropriate opportunity, or an existing customer asking for help before counting it toward a new-business goal.
Establish a baseline you can repeat
Choose a manageable set of questions drawn from actual sales conversations and customer decisions. Include service-fit questions, comparisons, and questions about your business by name. Keep those categories separate so a successful branded lookup does not inflate your assessment of unprompted recommendations.
For each captured answer, retain the full question, date, platform, mode, relevant location or account context, businesses named, and source links. Keep the original answer, not just a score assigned to it. Use the same definitions when another person reviews the next batch.
Report the number of matching observations alongside the total number of answers checked. That denominator describes your sample. It does not measure every customer query or your share of the whole market. Add new questions when the business needs them, while noting that the comparison has changed.
Our discussion of why ChatGPT may not recommend a business explains how to investigate an individual answer. The measurement task is to preserve those checks consistently enough to compare them over time.
Read Google's AI report within its actual scope
Google's Generative AI performance report (Search) reports impressions for AI Overviews and AI Mode. Its documentation says a missing report usually means insufficient impressions. Chart and page totals can differ because their aggregation differs. Review the current report documentation with your export.
Keep that report distinct from your overall Google Search performance. Do not rename all Search Console impressions as AI impressions, or turn impressions into a count of inquiries. Save the report name, property, dates, filters, and export so someone else can understand the comparison.
Google's Search generative AI control defaults to inclusion, and a child property can inherit its parent's setting. Check the actual property setting if eligibility is in question. A missing report alone does not establish that your site was excluded.
Use the available visibility data to choose what to examine. If a page receives impressions but its business purpose is unclear, review the page and its next step. That is a reason to investigate its usefulness. The impressions alone cannot explain why a reader did or did not contact you.
Connect attributable visits with website actions
GA4 includes an AI Assistant channel for traffic that meets its current classification rules. Google's AI Overviews and AI Mode are excluded from that channel; their non-ad traffic falls under Organic Search. Check the current channel definitions before grouping the traffic in your report.
The GA4 Traffic acquisition report includes Session source / medium, which describes the source and medium associated with a new session. Review that dimension alongside the channel grouping and relevant landing pages.
OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs from search results. Its publisher guidance explains this tracking detail. Inspect what your own analytics actually receives before creating the report.
Group visits using the source information your analytics captured. Leave unknown sources unknown. A customer might read an answer, remember the company, and contact it later through another route. Ask how the person found you, and keep that self-reported answer separate from the source captured by analytics.
Choose a meaningful website action, such as a completed inquiry or booking, and check that it is recorded correctly. Test the process through to the actual notification or appointment record. If booking happens on another domain, confirm how the handoff is measured before presenting a continuous journey.
A visit count and an inquiry count from the same month are not automatically connected. State the method used to associate an inquiry with its source, including the reporting window. Keep the original records available for review.
Let the sales team define a useful inquiry
Agree on a short set of qualification criteria with the people handling incoming requests. A contractor may need the project type and service location. A manufacturer may need the requested work and quantity. These are illustrative criteria; the actual business should set its own.
Use one inquiry record as the reference when someone submits a form, calls, and then books. Otherwise, three actions from the same opportunity could become three supposed leads. Separate tests, spam, duplicates, and requests outside your services before reporting new opportunities.
If someone becomes a customer, record that outcome and the date. Retain the attribution method that connects the customer to the original inquiry. Do not present all website revenue as AI-generated revenue because AI visibility improved during the same period.
A brief intake question can add context that a traffic report cannot provide: what prompted this person to contact the business? Keep the answer in the customer's own terms instead of choosing the source that makes the campaign look strongest.
Finish the report with the next decision
Hudson South distinguishes PUBLISHED, INDEXED, CITED, RANKED, and RECOMMENDED. Treat them as separate observations with dates and evidence. None should silently become a claim about leads or revenue.
If recommendations increase in your sample but identifiable visits remain limited, report both findings. If visits increase but inquiries do not, examine the page's relevance, offer, contact process, and measurement setup. If inquiries rise, check their fit and what happens after the first response.
Record other meaningful changes, including advertising, promotions, service changes, and revisions to tracking. A before-and-after comparison describes what happened. It cannot by itself isolate the contribution of one article or one channel.
For the wider content approach, start with getting your business found in AI search. Use this scorecard to decide which part of that work deserves attention next, with the measurement limits visible beside the results.
Make the report useful to the business
Book a 30-minute call with Hudson South to discuss your content goals, the evidence you have, and which questions your reporting needs to answer.
Book a 30-minute callSources
- Generative AI performance report (Search). Google Search Console Help. Read 2026-10-08.
- Search generative AI control. Google Search Console Help. Read 2026-10-08.
- Default channel group. Google Analytics Help. Read 2026-10-08.
- [GA4] Traffic acquisition report. Google Analytics Help. Read 2026-10-08.
- Publishers and Developers - FAQ. OpenAI. Read 2026-10-08.
