BIWEEKLYVerisca Editorial
AI marketing measurement needs a connected path, not one master number
A four-layer way to connect AI discovery with campaign outcomes and human choice.
Verisca Biweekly · September 15–28, 2026
Knowing whether an AI mentioned a brand is not enough to make the next marketing decision. Whether someone saw the mention, what explanation led to a page or action, and whether a person’s perception and choice changed are different questions.
The last two weeks did not produce one unified measurement system. They did show more tools breaking the journey into observable parts: image-led search, AI answers, conversations beside ads, carts and checkout. At the same time, observations continued to show that model and service changes can alter search behavior and citation counts, making a single metric difficult to read as durable performance.
This issue does not use article-count changes as a trend. Nine days of source material are missing from the comparison period, so the two periods do not have comparable collection volume. We focus on directions repeated across official announcements and comparative observations, along with the limits that remain.
What is emerging: AI touchpoints are extending from answers to action
On September 16, Google grouped several retail updates together: Merchant Center insights for brand and product discovery in AI Mode and AI Overviews, a U.S. beta of Business Agent for questions beside YouTube ads, and UCP capabilities linking carts and checkout. Markets and eligibility differ by feature. Google announcement ↗
On September 24, Google also announced Search Console reporting that distinguishes image-led searches through Lens, Circle to Search and image uploads. Data appears for sites receiving eligible traffic. Google Search Central announcement ↗
Together, these announcements do not establish that AI caused sales. They do expand the units marketers can observe:
- where a question or image led to brand discovery;
- which product facts and reasons to choose were explained;
- whether someone moved to a site, cart or conversation; and
- whether inquiry, purchase or brand perception changed afterward.
The campaign’s key message does not need to become a list of facts. If advertising creates a desirable meaning, the next touchpoint should connect that meaning with the customer’s situation and reliable product information. Marketing for AI and marketing for people meet where the same reason to choose survives across touchpoints—not where more information is simply added.
What persists: visibility and effect are not the same measure
During the comparison period, reporting highlighted the difficulty of reading Google Search Console’s AI-search impressions and positions like conventional rankings. The position of an AI answer block differs from the position of a link inside it, while an impression can under some conditions be counted without the link being seen. Search Engine Journal report ↗
A third-party comparison in the current period also found that a model change alone can alter search and citation patterns. Seer Interactive compared 346 matched prompts across six industries and reported that GPT-5.6 Luna used more domain-targeted searches than the prior model while consulting and citing fewer sources. This is one provider-run sample and does not represent all use. It does illustrate why fewer citations alone should not be treated as proof that a content strategy failed. Seer Interactive study ↗
Industry reporting points to the same unresolved issue: platforms favor different sources, answers vary, and shared measurement standards are still being developed. Digiday report ↗
New reporting expands the windows of observation. It does not combine discovery, explanation, movement and human choice into one number. What changed over these two weeks is not the completion of measurement, but a broader foundation for preserving different evidence at each stage of the path.
Separate four layers in the campaign you are running now
| Measurement layer | Question | Record | What this alone cannot establish |
|---|---|---|---|
| Discovery | In which question, image and service did the brand appear? | Service, model, market, prompt, mention and citation | Whether someone saw or preferred it |
| Explanation | For whom and in what situation was it recommended, and why? | Answer text, supporting source, factual error, recommendation condition | Whether brand perception changed |
| Movement | Where did the person go and what did they do? | Landing page, cart, conversation, click and completion | Whether AI caused the action |
| Human outcome | Did understanding, distinctiveness, preference or purchase change? | Survey, interview, experiment, inquiry and revenue | Whether AI caused every change |
Consider a hypothetical luggage campaign that proposes “a life where even an unplanned trip feels light.” One additional brand mention is a discovery change. An explanation that the bag suits someone carrying a laptop on a train is a recommendation-context change. Checking the storage layout and moving to an inquiry or purchase is a behavioral change. Whether customers remember the brand as a symbol of freer travel requires separate research with people.
Before combining the layers into a score, find the broken connection. Discovery may rise while the explanation remains generic. The explanation may improve while the landing page fails to carry the promise forward. Purchases may rise while price, promotion and seasonality also changed.
For the next two weeks, change one measurement sheet
Choose one active campaign and record these items together before buying another tool:
1. the perception the campaign intends to change and its one-sentence key message; 2. five questions real customers might ask, plus service, model, market and repeat conditions; 3. brand mention, recommendation reason, supporting source and factual error in each answer; 4. the connected page and the next action available to the customer; and 5. a way to examine human understanding and preference, or actual inquiries and purchases.
Do not create a total score in the first measurement. Record the break: “mentioned without a reason to choose,” “correct explanation but a different promise on the page,” or “movement observed, human attitude unmeasured.” Apply the next change only to that connection.
Scope and what remains unknown
The current period is September 15–28, 2026; the comparison period is September 1–14. The current input has 103 records and is missing September 22. The comparison input has nine records and is missing September 1–4 and 7–11. We therefore did not infer changes in article volume, topic share or market interest.
Official sources support the product direction; third-party comparative research and industry reporting support the measurement cautions. Selected product evidence is concentrated on Google, and we did not examine action and measurement features from other AI services under the same conditions. We did not measure a live campaign, Korean-account availability, customer perception, preference or revenue effect. The four layers and worksheet are an operating hypothesis to test, not a proven performance model.