A Google ranking, an AI citation and an enquiry are three different observations. Put them on the same slide and they look like a customer journey. Put them in the same sentence and a single verb can turn that resemblance into a claim nobody has proved.
TL;DR
- The verb in a performance report must not travel further than the evidence. Say "recorded", "appeared" or "displayed" until a method earns "drove".
- Each system has an evidential ceiling: Search Console, an AI capture and the CRM each permit a bounded statement and no more.
- Map the four-stage chain (appearance, access, response, qualification) and draw the missing links rather than colouring them in with prose.
- Split the report so each observation leads to its own decision. Unknown is an admissible result.
My rule is simple: the verb in a performance report must not travel further than the evidence. If Search Console recorded visibility, say the page appeared or received clicks within the recorded scope. If an AI answer displayed a citation, say exactly that. If the CRM recorded an enquiry, report the enquiry. Do not write drove, generated, influenced or contributed to until a defensible method connects the observations.
This sounds like a language problem. It is actually an evidence-chain design problem. The wording merely reveals where the chain breaks.
The most expensive word in the report is often the verb
Consider this sentence: The page ranked third, appeared in AI answers and drove eighteen enquiries. Every noun could point to a genuine record. The page may have an average position close to three in a saved Search Console view. A captured AI response may display a link to the page. The CRM may contain eighteen enquiries from the same reporting period. The sentence can still be false.
The shared subject, the page, makes the events feel sequential. The page ranked. An answer cited it. Buyers acted. But the sequence exists in the grammar unless a suitable identifier, referral, session record, buyer report or controlled analysis connects those events.
Co-occurrence belongs in a timeline; causation belongs in a method. Two events happening during the same month gives you permission to investigate a relationship. It does not give you permission to write the relationship as fact.
Swapping drove for helped does not repair the chain. Neither does supported, influenced, contributed to or the wonderfully foggy played a role in. Softer language can make the causal claim harder to challenge without making it better supported. I would rather present three bounded wins than one impressive fiction. That is not timid reporting. It gives the decision-maker three things they can actually inspect.
Give each record an evidential ceiling
Every measurement system has a ceiling: the strongest statement its retained data can support without borrowing meaning from another system. Google Search Console reports search performance within selected dimensions and periods. Its Performance report includes clicks, impressions, click-through rate and average position, with data that can be viewed by dimensions such as query, page, country, device, search appearance and date (Google Search Console Performance report). That can support a bounded statement: Search Console recorded the page’s clicks, impressions and average position in the saved UK Web view for the stated period. It cannot identify a named prospect behind an impression. It cannot prove that person later saw an AI response. It cannot connect a later telephone call to the earlier search exposure merely because both occurred.
Average position needs particular care. Google defines it as the average of the topmost position occupied by a property or page across recorded impressions. Results can also vary with circumstances such as time, location, device and search history (Google’s position methodology). “We ranked third” sounds like a fixed shelf position. “Search Console recorded an average position of 3.2 for this page, within this query set, country, device category, search type and period” describes an aggregate. The second sentence is less cinematic and far more useful.
An AI-answer capture has a different ceiling. If a saved ChatGPT response visibly cites a page, the capture can support a statement that the page appeared as a citation in that response under the recorded conditions. OpenAI explains that web-search responses may include citations and that a Sources panel can include cited sources and other relevant links. OpenAI also cautions that search output and citations can be incomplete, outdated or incorrect (OpenAI’s ChatGPT search guidance). The capture does not show how many other people received the same answer. It does not prove that anybody opened the citation, accepted the claim, remembered the brand or made contact.
A CRM record has another ceiling. It can establish that an enquiry was received, when it arrived, how it was classified and what happened next, provided those fields are actually retained. Without suitable attribution data, the CRM cannot identify the particular ranking or generated answer that caused it. Useful systems. Different permissions.
A citation is not a recommendation, and a mention is not a citation
AI visibility reports often compress several interface events into a single metric called “citations”. I would not approve that metric until the team defines what it counts. The observed event might be:
Those events are not interchangeable. They imply different relationships between the answer, entity and source. For each captured response, retain the question or prompt, response text, interface, visible placement, linked destination, date and any material account, location or personalisation conditions. Then use the narrowest description that the capture permits.
Even a genuine inline citation is not automatically a recommendation. It proves that the interface displayed a source relationship in that response. The surrounding answer may be positive, neutral, critical or simply factual. The source may support one sentence while the reader assumes it supports the whole answer. My preferred reporting question is not “Did we get cited?” It is “What role did the brand or page play in the answer?” That forces the analyst to inspect context rather than celebrate link-shaped pixels.
Map the Observation-to-Outcome Chain before claiming movement
I use four separate stages to test visibility claims. The names are deliberately plain because this is a control, not a marketing framework.
| Stage | Question | Suitable evidence | What it does not prove |
|---|---|---|---|
| Appearance | Was the page, brand or claim displayed? | Search Console record, saved result or captured AI response | That a particular buyer noticed it |
| Access | Did a person reach an owned destination? | Referral data, landing-page session, server record or another suitable visit record | That the visit changed preference |
| Response | Did the person take a meaningful action? | Form, call, booking, download or account event with defensible linkage | That the action was commercially suitable |
| Commercial qualification | Did the action become the kind of opportunity the business wants? | CRM disposition, sales qualification or transaction record | That the original appearance caused the entire outcome |
The chain is valuable because it shows exactly what is known, not because every organisation can join every stage. Buyers change devices. They return directly. They ask a colleague. They see a brand in one channel and convert through another. Privacy controls remove identifiers. Telephone enquiries arrive without a clean source. Some paths will remain unresolved after competent measurement. Unknown is an admissible result.
What matters is that a missing link remains visible. If the report records an appearance and a qualified opportunity but cannot observe access or response, draw the gap. Do not colour it in with prose.
Run the Verb Permission Test
Before a performance claim reaches a client, director or board, underline its consequential verbs. Then ask what retained evidence gives each verb permission to exist.
| Report verb | Minimum question to answer |
|---|---|
| appeared, displayed, recorded | Which system observed it, under what scope and when? |
| clicked, visited, arrived | Which event record connects the person or session to the destination? |
| enquired, booked, purchased | Which business record captures the action? |
| qualified | What explicit commercial definition was applied, and by whom or what process? |
| influenced, contributed, helped | What method establishes influence beyond timing or intuition? |
| drove, generated, caused | What evidence supports the causal connection and excludes plausible alternatives? |
The test is intentionally asymmetrical. Recorded carries a smaller burden than caused. A report should be confident at the level it has earned rather than vaguely cautious at every level. Here is the practical rewrite: Search Console recorded increased visibility and clicks for the page within the defined comparison. A saved AI response displayed an inline citation to the page under the captured conditions. The CRM recorded eighteen enquiries during the reporting period. The available records do not connect those enquiries to the search observation or saved response.
That version does not demote anything. It preserves the ranking evidence, the citation and the enquiries. It removes only the journey nobody observed. If additional linkage exists, add it precisely. A recorded referral and landing-page session may connect a source to a visit. A form submission tied to that session may extend the observed journey. A CRM disposition may establish whether the enquiry met the commercial definition. Each additional record raises the evidential ceiling. None grants unlimited permission.
Want a second reader for the next visibility report before it reaches the board? We underline the verbs and show which ones the evidence has earned.
Split the report so people can make better decisions
A blended success sentence feels efficient, but it hides the next action. Separating the observations makes the report operational. If search visibility changed, inspect the affected queries, pages, devices, countries and search appearances. Decide whether the change matches the intended audience and topic, not merely whether the graph points upwards. If an AI response cited the business, inspect the question, wording, accuracy, context and destination. Decide whether the answer represents the brand correctly and whether the cited page deserves to carry that claim. If enquiry volume changed, inspect source capture, qualification, buyer fit and sales disposition. Decide whether the business received more of the opportunities it wants or simply more activity.
If the organisation genuinely needs attribution, decide how much uncertainty matters and what a proportionate investigation would cost. Better instrumentation may narrow the gap. It may not eliminate it. This is where my view becomes unfashionable: I do not think marketing reporting improves when every visibility observation is forced to end in revenue. It improves when each observation leads to the right decision. A citation-quality problem, landing-page problem and sales-qualification problem require different responses. One triumphant number can conceal all three.
Keep the chain honest without making the report useless
Evidence discipline can become theatre of its own. A report that refuses to interpret anything until it has experimental proof will be technically cautious and commercially inert. The answer is not to ban judgement. It is to label its role. State observations directly. Present plausible explanations as hypotheses. Make recommendations as professional judgement. Reserve causal language for methods that can carry it. When evidence conflicts, show the conflict. When a useful inference remains uncertain, explain what would raise or lower confidence.
I would approve this sentence: The timing and query pattern make the visibility change a plausible contributor to increased enquiries, but the available records do not identify the affected buyers. I would inspect landing-page and CRM data before using the change to forecast demand. It makes an inference, preserves its limitation and still produces an action. It does not hide behind “anything is possible”, and it does not pretend the buyer journey has been reconstructed.
Before the next visibility report leaves the room, find every sentence that travels from appearance to money. Mark the stages the evidence actually records. Circle the missing links. Then choose one of three honest actions: investigate the gap, report the observations separately, or decide that stronger attribution is not worth the cost.
Preserve the ranking. Preserve the citation. Preserve the enquiry. Just do not ask the verb to manufacture the customer journey between them.
Good to know
FAQs
No. Search Console records clicks, impressions, CTR and average position within selected dimensions and periods. It cannot identify a named prospect behind an impression or connect a later call or form to the earlier search exposure. That linkage needs a referral, session or attribution record.
Define it before reporting it. An inline citation, a sources-panel link, an unlinked mention, a related result and unattributed text that resembles your page are different events with different relationships to the answer. Use the narrowest description the capture permits, and inspect the answer’s context rather than counting links.
Softer verbs make the causal claim harder to challenge without making it better supported. The Verb Permission Test asks what retained evidence gives each verb permission to exist. “Recorded” carries a smaller burden than “caused”; be confident at the level you have earned.
Unknown is an admissible result. Draw the missing link rather than colouring it in with prose, report the observations separately so each leads to its own decision, and decide whether stronger attribution is worth the cost of better instrumentation.
Next step
Does your report earn its verbs?
We map the observation-to-outcome chain across Search Console, AI captures and your CRM, mark what the evidence records and separate each observation into a decision you can act on.