An AI answer can cite your studio without sending anyone to your portfolio. A visitor can arrive from an AI platform without telling your analytics which question they asked. Neither observation, by itself, tells you whether someone sent a useful project brief.
Those differences should determine how you report performance and where you spend next. Citation monitoring asks whether your pages appear in a defined sample of answers. Referral reporting asks which recorded sessions are attributed to an identifiable source. Enquiry reporting asks whether visitors take a commercially useful next step.
We approach AI referral measurement as a way to separate those decisions, not combine them into one visibility score. For a creative business, being cited for a broad definition is different from helping a buyer shortlist a production partner. Your report needs to preserve that difference before it can guide work on your pages.
01
Start With The Quantity You Actually Need
An observed AI referral session is a recorded website session attributed to a source you have identified as an AI platform. Its meaning depends on the analytics dimensions and classification rules you use. It measures recorded arrival at your site, not every occasion on which an AI answer mentions you.
A citation observation is a record of a source appearing in an answer you inspected. Define what qualifies before collecting observations. A linked portfolio page, an unlinked studio name and a link to a third-party directory are different things. If the measure is citations to your own site, a directory’s description of your studio should not silently count as one.
Citation share needs an explicit denominator. You might measure your site’s qualifying citations as a share of all qualifying citations across sampled answers. Alternatively, you might measure the proportion of inspected answers that cite your site. Those measures answer different questions: the former describes your presence among cited sources; the latter describes how widely you appear across the sample. Name them accordingly rather than calling both “AI visibility”.
An enquiry belongs in a separate record. Decide whether you mean a completed contact form, a request for availability or a qualified brief that matches your services. A click on a contact button is not equivalent to a submitted enquiry, and a submitted enquiry is not necessarily a suitable project. That definition keeps attention on the work you want rather than the easiest event to count.
02
Keep Referral Reporting At Session Scope
Google distinguishes user, session and event scopes in its traffic-source reporting. For a report about arriving sessions, use session-scoped source information consistently. First-user source addresses acquisition of the user, while event-scoped attribution addresses a different reporting question. Mixing these scopes can make apparently comparable figures describe different parts of a visitor’s history.
The Google Analytics traffic-source documentation states: “Each session is associated with only one campaign or traffic source”. It also explains that new campaign or traffic-source values collected during an existing session do not replace that session’s source, although they can be associated with events.
The practical consequence is that a session-source report is not a complete sequence of everything someone consulted. A prospective client may research a studio, inspect projects and return before making contact. The source assigned to a recorded session cannot, on its own, explain that whole decision.
For a repeatable referral report, record the date range, session source and medium, landing page, and the exact source values included in your AI grouping. Keep the grouping available for inspection. If you add another source later, note the change or reprocess the comparison period using the same rule. Otherwise, an apparent increase may partly reflect a broader definition.
Landing pages make the result more useful. Sessions reaching a description of your production services raise different questions from sessions reaching a journal article or an old release announcement. Check whether each destination gives the visitor a sensible route towards the relevant portfolio, collection or enquiry page. Source totals alone cannot show whether the arrival page serves the visitor’s likely purpose.
03
Missing Referrers Leave A Specific Gap
Analytics relies on collected campaign and traffic-source information. Google documents that a session can be processed as direct traffic when referral-source information is unavailable, or when the referring source or search term has been configured to be ignored.
That means identifiable AI referrals are not a complete count of every visit influenced by AI. It does not mean the direct category should be relabelled as hidden AI traffic. Without identifying evidence, the origin remains unresolved. Assigning those sessions to AI would replace a measurement gap with an unsupported claim.
Keep this limitation close to the referral figure: the report covers sessions attributed to the sources in your defined grouping. This tells a decision maker what is included without suggesting that the data is useless.
Google Search requires another scope distinction. Google’s AI search guidance says that sites appearing in AI Overviews and AI Mode are included in overall Search Console traffic, specifically in the Performance report’s “Web” search type. A total from that documented reporting scope therefore includes more than AI-feature activity; it should not be presented as an isolated AI referral total.
This is a statement about that particular reporting scope, not a claim that dedicated AI reporting cannot exist. When evaluating another report, inspect its definition, coverage and collection method. The label on a dashboard is not enough to establish what its figures represent.
04
Build A Citation Sample You Can Repeat
Citation monitoring becomes useful when you can explain why the questions were chosen and repeat the conditions. The following is a proposed measurement method, not a study we have run.
Start with a fixed set of questions tied to decisions your buyers make. For a production studio, those could concern service suitability, relevant project experience or a location requirement. Separate broad research questions from supplier-selection questions. Otherwise, frequent appearances in introductory answers can obscure absence from the questions closest to an enquiry.
Avoid selecting only questions where you already appear. That creates a favourable collection of examples rather than a useful baseline. Give every question a stable identifier and preserve its exact wording. Keep branded questions separate from unbranded discovery questions: asking about a named studio tests something different from asking which studio might suit a project.
For each observation, record:
- The exact question and its identifier.
- The platform, feature or mode, and any displayed model information.
- The date, time, location setting and login state.
- Whether it was a fresh conversation or included prior context.
- The answer and qualifying source links.
- Whether your site appeared, which page appeared and how the citation was used.
- Any access error, incomplete response or other collection failure.
These details make comparison interpretable. If the platform or settings change, mark the break rather than treating the new observations as a seamless continuation. Google notes that AI Overviews and AI Mode can use different models and techniques, and that their responses and links vary. Pooling their results without keeping the feature identifiable would hide a potentially important difference.
Preserve enough of the answer to examine the citation’s role. A project page used to support a statement about your work is different from a homepage appearing in a list with little explanation. Both may qualify under your counting rule, but the surrounding answer helps you decide whether your services are being described accurately.
05
Publish The Denominator And The Failures
Set counting rules before looking at the results. If you count source links, decide whether repeated links to the same domain within an answer count separately. If you count answers containing your site, count each answer consistently regardless of how many of your pages it links to.
For citation share among sources, the denominator is all qualifying citations in the defined sample. For answer coverage, it is all successfully inspected answers under the chosen rule. Report the sample’s scope alongside either result. Neither measure establishes your share of every answer a platform gives to every user.
Failures must remain visible. A request that could not be completed is not an observed answer without a citation. An answer that completed successfully but contained no qualifying sources is a different outcome. Keep those categories separate so that collection problems do not look like lost visibility.
For Google AI Overviews, also distinguish a search where no Overview appeared from an Overview that appeared without citing your site. Google says Overviews often do not trigger. Whether non-triggering searches belong in a particular denominator depends on the question you are measuring, but the rule must be explicit and unchanged across comparisons.
Decide in advance how to handle retries and repeat runs. Retrying until your page appears would bias the sample. If repeat observations are part of the method, apply that method consistently and retain the different outcomes. A changed sample can still be useful, but it starts a different comparison unless you preserve a common subset.
06
Read The Measures Together Without Inventing A Funnel
Citation observations, referrals and enquiries can sit beside each other in a report. They should not automatically become successive stages of a measured funnel.
A sampled answer is not an impression count for the platform’s audience. Your analytics sessions usually do not identify the exact sampled answer that preceded them. Dividing referrals by sampled citations would therefore not produce a defensible click-through rate: the quantities do not describe a matched population.
Instead, use each measure to choose an investigation. If citations increase while recorded referrals remain flat, inspect the questions and cited pages before expanding production. Are the answers addressing buyer decisions, or questions that may be resolved without visiting your site? The observation suggests what to examine; it does not establish why people did not click.
If identifiable referrals reach your portfolio but enquiries remain weak, inspect the destination and enquiry path. Can visitors understand your role in the projects, the service you provide and how to discuss a suitable commission? A portfolio can demonstrate visual quality while leaving those buying questions unanswered.
If enquiries increase, retain the distinction between recorded attribution and a client’s account of discovery. A prospect saying they found you through an AI answer is useful qualitative evidence. It is not a reason to overwrite an unrelated session source. Keeping both records gives you more context without making either claim more than it shows.
07
Choose The Work That Matches The Evidence
A citation gap is a reason to inspect the pages that should answer the sampled questions, not automatically commission more articles. Our AI search optimisation work examines technical access, entity clarity, useful answers, visible proof and competitor citations. For a creative studio, that can mean clarifying who delivered a project, what service the work demonstrates and why the page supports a buyer’s comparison.
When the problem spans discovery, landing pages and enquiries, our SEO audits connect crawlability, content, internal links and conversion paths to a prioritised roadmap. That is the more relevant starting point when you cannot yet tell whether the obstacle is access, missing service detail or an unclear next step.
The practical answer is to keep separate measures with separate definitions: recorded AI-attributed sessions for arrivals, a fixed citation sample for observed answer presence, and defined enquiries for commercial response. Use the relationship between them to select the next check, not to claim a connection the records cannot establish.
For a review, bring the source-grouping rules, sampled questions and priority landing pages alongside the enquiry definition. Those give us a concrete basis for examining whether buyers can find, understand and act on your work, rather than treating a higher citation count as the finished result.