# How to Measure Phone Enquiries After AI Discovery By AI Visibility editorial team AI-assisted article. A customer asks an AI assistant for a local service, sees a phone number and dials it. The business may never see a website session. If the customer instead taps a call button on a business profile, the profile can record the tap without establishing that anyone answered. The phone team can supply a different kind of evidence by recording the enquiry and asking how the caller first found the business. Measure the phone journey in two parts. Count genuine enquiries from the phone team's own records, then show how many callers reported AI discovery and how often staff obtained a usable answer. That lets a Dubai business track a channel that may leave no website trail, while keeping the size of the evidence gap visible. Decide what counts as one phone enquiry Start with the decision the report supports. A receptionist planning staffing needs to count ringing calls and missed calls. A sales manager evaluating new business needs to count distinct enquiries. A marketing analyst investigating discovery needs a source response linked to those enquiries. The same phone call can appear in all three views, but the counts serve different decisions. Write an operational definition. For example, count a new request for an appointment or quote once it reaches a staff member and the staff member can identify the service the caller needs. Exclude wrong numbers, spam, supplier calls and routine requests about existing jobs from that particular report. Retain those outcomes in the call ledger so the team can reconcile why the enquiry total is lower than the connection total. Decide how follow-ups work. A customer might ring the main line, then call a branch to confirm the same quote. Link the second contact to the existing enquiry when staff have a reliable reference. If a callback converts an unanswered call into a new enquiry, connect it to the earlier attempt and record the time the enquiry was first established. If a customer raises a separate service request, use the business's definition of a new enquiry. Keep the original call records even when several point to one enquiry. When branches share a switchboard, distinguish the number dialed, the team that answered and the location the customer wants to visit. A chain-level discovery report may combine the enquiries. A branch report needs the intended location and an explicit rule for calls transferred between teams. A transfer can add a call leg without adding a customer or an enquiry. Ask a neutral question during the conversation Once the immediate reason for calling is clear and there is room in the conversation, ask, “How did you first hear about us?” Let the caller answer in their own words. A leading question such as “Did you find us on ChatGPT?” can change what people say and leaves the team unsure whether the platform would have been mentioned without prompting. Use a separate question if the route to the number matters, such as “Where did you find the number you called today?” Someone may have first heard of the business from an AI assistant and then looked up the number on Maps. Someone else may know the business through a colleague and use an assistant to check the number. The source at first discovery and the route immediately before the call can differ. The report can use both fields and label them separately. Use the same intent across languages. Agree on the English and Arabic versions staff will use, and check that both ask about first discovery rather than asking where the caller got the number. A caller who switches languages or gives a mixed-language platform name can still provide a usable answer. Train staff to capture the source in the same way. AAPOR's survey guidance explains the tradeoff between open and closed questions. Here an open first question is useful because the caller can name a route the business did not anticipate. AAPOR best practices for survey research (https://aapor.org/standards-and-ethics/best-practices/) If the caller says “online” or “Google,” one neutral clarification can help when the conversation allows. Ask what they remember seeing. If they can describe an assistant answer but cannot name a provider, code the response at the level your evidence supports. If the description stays ambiguous, retain it as unclear. A quick phone interaction may end before a follow-up question is appropriate. Record that outcome separately from a caller who explicitly does not remember. Give staff a short source record Use fields that staff can fill while handling a real enquiry. Keep the enquiry reference, call date, service and branch, whether the source question was asked, and the response category. A brief source-only note can explain ambiguous cases or record two different touchpoints. Keep the operational record in the team's approved system and keep unnecessary personal details out of the measurement export. Useful response categories include named assistant, assistant described without a reliable platform name, known other source, unclear response, declined, and not asked. Write a definition for each. For example, your AI-reported total could include a named assistant or a sufficiently specific description of an assistant answer. It would leave generic “Google” responses in their original category unless further evidence supports recoding. The rule should also say how you handle multiple sources. For first discovery, preserve the first source the caller identified and add the later route in a separate field. The official The Sewing Corner Instagram post below illustrates a location-finding touchpoint. A customer could discover a shop through an assistant and use a social map graphic to decide which branch to call. The post establishes what the business published. A caller's account or a separate record is needed to establish whether that journey occurred. When staff record both discovery and the route to the number, a report can retain the distinction without choosing one source for both. View the selected official Instagram post (https://www.instagram.com/thesewingcorner/p/DeyZzowET7c/) Official The Sewing Corner branch-finding graphic. The first carousel slide was previously verified. Keep call-button metrics beside the phone ledger Google Business Profile defines its Calls performance metric as clicks on the call button. A click can precede a call, but it does not establish that the phone rang, that someone answered, or that the caller made a new enquiry. Report profile call-button clicks as a platform activity count. Report connected calls and accepted enquiries from the records that support those outcomes. Google Business Profile performance metrics (https://support.google.com/business/answer/9918094) Google Ads call reporting provides another illustration. Its documentation distinguishes calls from call conversions and explains that a call may fail to reach the duration threshold used to count a conversion. It also notes that call and conversion dates can differ under the product's reporting method. Those definitions belong to Google Ads. A local business still needs its own record of what callers asked and where they said they discovered the business. Google Ads call reporting definitions (https://support.google.com/google-ads/answer/7180997?hl=en) Where a website measures taps on phone links, keep those events separate as well. A measured AI session can include a phone-link tap and the caller can still abandon the dialer. A call may also arrive without a tap if the person copies or types the number. If the team has an approved and reliable way to reconcile a measured route with a call record, retain that linkage as a separate evidence field. A similar time and a matching branch are useful clues for an investigation and can fit several callers. Reconcile the source numbers before reporting The following figures are constructed. A team has thirty inbound attempts during its reporting window. Staff answer twenty-two as they arrive and connect with four callers later by returning missed calls. The twenty-six conversations include six outside the enquiry definition and four repeat contacts. That leaves sixteen distinct new phone enquiries in the fixture. Stage Count What the number describes Inbound call attempts 30 Includes eight initially missed calls Answered inbound calls 22 Connections when the inbound calls arrived Connected callbacks 4 Returned calls that reached the customer Connected conversations reviewed 26 Answered inbound calls plus connected callbacks Distinct new phone enquiries 16 Excludes six out-of-scope conversations and four repeats New enquiries asked about discovery 12 Four were not asked Usable discovery responses 9 Three asked enquiries remained unresolved AI discovery reported 3 A subset of the nine usable responses Report that three of sixteen new enquiries included reported AI discovery, and show that staff asked twelve of sixteen. If you also report that three of nine usable responses named AI, keep the denominator in the same line. The remaining seven enquiries consist of three unresolved responses and four cases where staff did not ask. The source mix of the nine known replies cannot tell you the source mix of those seven. Track changes in collection alongside changes in the headline number. If a new script increases the number of callers asked, recorded AI discovery can rise while underlying demand stays steady. A branch with lower coverage may simply have more calls coming in during a shift when staff rarely ask. Compare response completeness by branch, time or team when those differences affect the decision. Keep the call counts visible so a handful of responses does not look like a substantial trend. Keep the process consistent across shifts Start with one owner who can review the call dispositions and the discovery field. Give staff examples of how to handle “someone sent me your number,” “Google told me,” or “I asked ChatGPT and then found you on Maps.” Review a few records together and record how the team resolved ambiguous examples. This is especially useful when one branch captures the first discovery source and another has been recording the last place the caller saw the phone number. After the first reporting period, check which shifts and branches have missing source records. Some gaps will reflect short calls, transfers or urgent service requests. A callback script can improve coverage when a staff member has time after dealing with the customer's reason for calling. Keep the source question optional in the workflow and let the enquiry proceed even when no source can be recorded. Use a consistent period boundary. A missed call on the final evening of the month may become a genuine enquiry after a callback the next day. The call-attempt report can keep the original date while the enquiry report follows the date the business first established the new request. Document the rule and avoid moving dates simply to align the two totals. If a report is revised to include late records, preserve the earlier cutoff so readers can explain the change. Put the phone result in the wider enquiry report Compare phone enquiries with forms and messages at the level of distinct accepted requests. If the same customer submits a form and calls about that request, link the records when the approved workflow makes the match reliable. Retain the separate contact events. That preserves the operational workload while keeping the enquiry total from counting one request twice. A monthly line can read, “Sixteen new phone enquiries, twelve asked about discovery, nine usable answers, and three callers reporting AI discovery.” Add separately measured platform taps if the reader needs the operational context. The existing AI Visibility referral-to-enquiry guide covers how to bring forms, calls and WhatsApp together. This phone workflow supplies the disposition and source-coverage evidence that makes the phone line reviewable. AI Visibility referral-to-enquiry guide (/article/how-do-you-measure-whether-ai-search-brought-you-anything-from-ai-referral-to-enquiry/) Phone enquiry measurement record Keep the reporting dates, branch scope, call and callback definitions, new enquiry rule, question wording, source codes and duplicate handling. Publish the distinct enquiry count, asked count, usable response count and AI-reported count. Record the evidence behind any verified call linkage separately from caller-reported discovery. The team can then tell whether the next month brought more enquiries, more complete source collection, or both. AI Visibility is published by Lunasol, which offers GEO and AEO services. The phone ledger figures are constructed to illustrate the reconciliation. About AI Visibility (https://aivisibility.ae/about/)