Last issue, I wrote about the trust you need before leadership will fund something they can't trace neatly to a conversion. I want to spend this issue on what comes next, and that’s what this issue is dedicated to.
A client recently asked whether they should focus their AI search effort on the top, middle, or bottom of the funnel. My recommendation was to start with the buying questions closest to revenue.
But before choosing what to track, we needed to agree on what the work was supposed to accomplish.
So this is the measurement edition! Same sections, one focus: what we can observe, what we're totally guessing at, and what would convince us the work is paying off.
And yes, measurement is almost completely different than traditional SEO and requires different methods of capture:
However good a dashboard can give you the numbers, it still requires someone or a team to decide the value of each, so let’s get into it.
Onward!
What are we actually trying to improve??
I break AI search measurement into 3 layers. Each should help someone make a decision and have some sort of strategy attached.
First, AI Readiness
Can AI access your important pages, and does the content answer the questions buyers have? Think about this from a technical perspective, follow me on this analogy: can people use streets to get to your house? Is the road clear, the door unlocked? Are your rooms furnished so you can tell what each is?
This is the most technical part of AI Search and is a prerequisite to get the later / more mature stages.
Page accessibility: How many priority pages can the relevant AI search crawlers access?
Content coverage: How many important buyer questions does your content answer?
Information accuracy: How many reviewed pages have accurate, current information?
Truly what we’re looking to do is open up your owned properties to crawlers, give them the information they are looking for, and become eligible.
Next, Competitive Visibility
Where does your brand appear in the answers buyers could encounter, and what do those answers say about you?
This is largely reliant on prompt measurement, which includes capturing all the criteria of an answer over time to help determine how visible you are in answer engines. Prompt tracking helps you see how often you appear, how you’re described, and how you compare with competitors across the questions you monitor. That gives you a view of competitive visibility within your sample.
Mention rate: How often your brand appears.
Recommendation rate: How often your brand is suggested as an option.
Citation rate: How often your website is used as a source.
Share of mentions: How your presence compares with tracked competitors.
Answer accuracy: How often your brand is described correctly.
Keep those measures separate. A citation means a source was used. A recommendation means the brand was put forward but neither will tell if someone clicked on what your offering.
Separate branded questions from unbranded discovery, too. Keep the tracked questions, platforms, and dates consistent, and make clear what each percentage counts. This stuff matters a lot, and if set up incorrectly, it will be the “tail wagging the dog”. Also, just focusing on platform prompts or a visibility score is not a strategy, it’s pleasing a system :)
If you need help with choosing software, just ask.
End boss: Business Impact
Are we seeing useful visits, qualified conversations, and business connected to AI search? At the end of the day, this is what the entire industry is gunning for so working towards solving this is extremely important.
AI referral visits and leads: Traffic and inquiries you can identify as coming from AI.
AI referral conversion rate: How often those visits become qualified leads or customers.
Buyer-reported influence: How many buyers say AI helped them find or evaluate you.
Associated pipeline and revenue: Opportunities and closed sales connected to those signals, reported separately.
Estimated incremental lift: Additional business a credible comparison test suggests the work produced.
Keep tracked conversions and buyer-reported influence distinct, and count an opportunity once if it appears in both. A buyer saying ChatGPT helped them find you is useful evidence. It doesn’t prove the sale wouldn’t have happened otherwise.
You don’t need every metric here. Pick the ones that help your team decide what to fix, where to focus, or whether to keep investing.
Attribution and measurement are getting more and more difficult, which I’ll dive into a bit later.
Your brand name can bring up more than your website
Chris Long reported that more AI Overviews are happening on branded searches, along with a lot of other great researchers.
Third-party, or god-forbid a competitor’s descriptions can sit right inside the answer someone gets when they search for you.
My concern is what happens when that includes your dirty laundry. An old complaint, an outdated product claim, a bad experience someone wrote about. More visibility doesn't automatically mean a better introduction!
These are risks to check for, not a claim that every branded answer is negative. But a report that counts your name and stops there can miss the thing your audience reads.
Check brand-only searches separately from reviews, comparisons, and complaints. Do things like read the answer, open the cited sources, and check ‘em out. Is the information accurate and current? Does the source support what the answer says?
A negative answer and a false answer need different responses. If the complaint is legit, the work may belong with the team responsible for CX. If the information is wrong, document the error and follow the trail.
Either way, somebody needs to own the next action cause some sentiment chart won't handle that conversation for you.
Your ChatGPT traffic may have lost its label
Cedric Huesler's analysis reports missing source tags and referrer information on ChatGPT links.
Across about 500 pages where ChatGPT previously supplied at least 5% of visits, he compared October 9–13 with the preceding week. Attributed ChatGPT visits fell by roughly 31k a day while Direct rose by roughly 26k on those pages. Wow.
He calls the evidence early and sampled, and updated his post to say this may be a bug, which matters.
Before telling leadership AI stopped sending buyers, check whether the visits stopped carrying a source label. Google's explanation of Direct traffic includes missing source information. Compare affected pages and devices, test the links and redirects, and annotate the reporting break.
There’s 1 horrible thing you have to watch out for, potential reporting bugs.
Branded search is getting a different front door
Ryan Law and Xibeijia Guan at Ahrefs found AI Overviews in about 83% of their US desktop branded-keyword search results on September 29, 2026. Their July–September analysis covered 232.8 million search-result crawls, including 92.5 million branded-keyword crawls.
Those are crawls, not unique people. Still, it reinforces Chris’s point: check what appears when someone already knows your name. More appearances can come with more third-party descriptions to review.
A citation drop can begin outside your company
We talked about the cratering of Reddit’s citation’s in August. Separately, Reddit has documented restrictions on unknown bots and crawlers. Source access is a real dependency, and when crazy things like this happen, citation patterns will fluctuate.
Your chart can move because a platform changed, a source became harder to access, or the measurement changed - or hell, even the circular finance and partnership deals happening could be expiring (like Google & Reddit).
My take: rigorous measurement means keeping track of how the evidence was collected, including when the method stops being comparable. These systems are still changing underneath us.
Don’t hold your breath that it will all be worked out soon.
This brings several signals together: identifiable referrals, relationships between AI Overviews and landing-page performance, Direct traffic, branded search demand, and what buyers tell us.
I'd keep those in the conversation but I'd also label the limits. Several lines moving together can help you choose what to investigate and they don't establish incremental revenue.
The assignment this week: get marketing, sales and RevOps to agree on one test.
Choose one buying question and one change. Improving the evidence about switching providers is specific enough to investigate. “Improve our AI visibility” is so freakin’ vague.
Name the business outcome and the decision. Agree on what result would justify continuing or expanding. Pick a time horizon that fits the sales cycle and enough volume to learn something (B2B/B2C matters).
Set the baseline and comparison. Randomize where feasible or explain why the comparison group is credible, check prior trends, and account for other campaigns and spillover.
Keep a change log. Content, campaigns, tagging, prompt sets, and platform shifts all belong in it.
Keep buyers' own words alongside the CRM category. Track how often you collect an answer, deduplicate overlapping opportunities, and separate pipe from c/w revenue.
Give someone ownership of the review. If you're an agency, agree which parts your team owns and which require the client's sales, technical, or RevOps teams. Otherwise, all this report will do is identify a problem nobody will fix.
How do I know this newly created business we wouldn't have won anyway?
Welcome to incrementality testing, and this is the question I'd put at the center of the measurement conversation.
If a buyer says ChatGPT helped them find you, keep that information. Connect it to the opportunity, cause it tells you something useful about influence.
It doesn't tell you whether that buyer would have found you anyway.
Incrementality asks how much extra business the work produced compared with what would have happened without it. We don't get to watch the same buyer live both versions, so the comparison has to do some serious work.
So, when revenue is the priority (duh), I want to start with the questions people ask while evaluating an approach, comparing providers or deciding whether a purchase makes sense.
Let’s take a hypothetical managed IT provider. The team improves a page explaining how to switch providers without disrupting operations. It adds evidence, answers real objections and gives a buyer something useful to take into an internal conversation.
I'd look for a credible comparison we can leave unchanged, check whether the groups behaved similarly before the work, and allow time for the buying cycle.
Suppose qualified opportunities improve, great. Even a credible test may tell us the content helped across search and sales. Separating the AI contribution is another claim that needs evidence. We shouldn't give AI exclusive credit for a page that also ranks in Google and gets used by the sales team.
And if the business doesn't have enough volume or a credible comparison? I'd report the changes we observed, what buyers told us, and what remains uncertain. Then choose the next useful test.
Welp.
Less satisfying than one clean ROI number, but hopefully this brings some more clarity to an unstructured measurement world :)
If you need help with measurement or anything else relating to AI search, let’s connect, I have an open spot on my roster coming up.
Until next time,
Andrew