Google’s generative AI performance report tells you how many times your pages appeared inside an AI answer. It does not tell you how many people clicked, because clicks are not in the report at all.
That is the honest starting point for anyone being asked what answer engine visibility returned last quarter, and it is why most attempts to answer that question quietly fall apart. What follows is what can be measured, what cannot, and how to hold the conversation without overstating either.
You cannot prove AEO and GEO return the way a partner usually means it, because Google’s own report contains impressions and no clicks. What you can show is a defensible set of leading indicators and an honest read on scale.
Presented as a positioning investment with a review date, that argument survives scrutiny. Presented as an acquisition channel, it does not.
The Report Google Gives You Has No Clicks in It
Search Console added a generative AI performance report, and Google’s documentation is specific about what the numbers are. Impressions are “how many times links to your site were shown to a user in a generative AI feature on Google Search.” That is the whole metric.
Three limitations matter for anyone building a case from it. The report combines AI Overviews and AI Mode without separating them, so you cannot tell which surface produced what. Google states that “not all properties have access to the report, as we’re rolling out over time,” so a flat line may mean no access rather than no visibility. And a site needs enough impressions before any data appears at all.
None of that makes the report useless. It makes it a visibility measure rather than a performance measure, and the difference is the entire argument.
At Current Scale This Is Not a Traffic Channel
The marketing agency Amsive, which sells the services this finding could have supported, published an analysis of 54 websites covering the most recent six months of data for each. It found organic traffic converting at 4.60% and LLM referral traffic at 4.87%, a gap of 0.27 percentage points. A paired t-test put it at p = 0.794, which is to say the difference was not statistically significant.
The scale figure is the one to bring to a partner. On nearly 90% of those sites, LLM referrals accounted for less than 0.6% of total traffic. Organic, over the same set, produced roughly a third of sessions and a third of conversions.
Read together, those numbers say something specific. AI referral traffic is not better traffic, and there is not much of it yet. A firm funding AEO and GEO in 2026 is not buying a measurable acquisition channel this year, and any case built on the premise that it is will not survive the first person who checks.
What You Can Actually Show a Partner
Four things are observable, and none of them requires a claim the data will not support. Establishing where a firm currently stands across them is what a baseline audit produces.
Generative AI impressions over time, from Search Console, framed as how often the firm appears inside AI answers rather than how much traffic it produced. Branded search volume, because a person who reads a firm’s name in an AI answer and later searches for it directly is the most common path this channel produces. Direct traffic, watched for the same reason and with the same caution. And what intake hears, which is the only place a human will actually say where they heard the name.
Track them as a set, monthly, alongside a fifth column nobody automates: which prompts your firm appears in. Running a fixed list of client questions through the engines each month and recording who gets cited is the closest thing to a scoreboard this channel has, and it is the method behind our own employment law citation audit.
None of that is proof in the sense a partner means. The closest thing to proof is a firm that stayed with it, which is why the more useful thing to bring is one firm’s results over a year rather than a projection.
How to Have the Conversation
Lead with the limitation. A partner who hears “clicks are not in the report” from you will not later discover it and wonder what else was oversold.
Then name what the investment actually is, with what this work costs on the other side of it. Answer engine visibility is a positioning bet, closer to being the firm a referral source mentions than to buying leads. It shows up in whether the firm is present when someone asks an assistant about their problem, and its effect on intake arrives on a lag.
Set a review date and say out loud what would falsify the bet. If impressions are flat after two quarters of consistent publishing, if the firm appears in none of its tracked prompts, and if intake never hears the channel mentioned, that is a real answer and it argues for stopping. A recommendation that cannot fail is not a recommendation.
What Not to Bring Into the Room
The statistics circulating about AI traffic converting several times better than organic mostly trace back to a single company’s own funnel, in one industry, over thirty days. The 54-site analysis above found parity, not a multiple. Where a number matters, Google’s own AI guidance is usually a better place to start than a vendor blog.
Cost-per-lead benchmarks for legal are worth the same caution. Most of the widely quoted ranges come from marketing agencies publishing their own estimates without disclosed method or sample, and a partner who checks one and finds nothing behind it will discount everything else you said. We have written separately about the claims that do not survive checking.
The case for answer engine optimization is strong enough on observable evidence. It does not need a number that will not survive being looked up.
Frequently Asked Questions
Can you track AI search traffic in Google Analytics?
Partly. Referrals that carry a source can be segmented, but answers where the reader never clicks leave no trace, and those are most of them. Treat analytics as a floor, not a count.
How long before a firm should expect to see anything?
Give it two quarters of consistent publishing before judging, then judge honestly against impressions, tracked prompts and intake. Set that date at the start so the review is not a negotiation.
Is AEO worth funding if the traffic is under one percent?
That depends on whether being present in AI answers matters to your market. The traffic argument is weak today. The positioning argument is the real one, and it should be made on its own terms.
What is the single best number to report monthly?
The share of your tracked client questions where the firm gets cited. It moves before traffic does, it is comparable against competitors, and it is the one figure a partner can interpret without training.
Build the Review Before the Meeting
The four measures above take about an hour a month to maintain once somebody owns them. The work is the first version: choosing the prompts that match your practice areas, establishing where the firm currently stands, and writing the numbers so a partner reads them the way you meant them.
That baseline is the first thing we build for a firm, before any content is written, usually alongside an audit of the pages you already have. A licensed attorney reviews everything we deliver, this included.
Ask for an AEO baseline on 877-486-8123. Or send over the objection you expect to hear, and we will tell you whether the evidence supports it.
David Arato, JD is the founder of Lexicon Legal Content, an attorney-owned legal content marketing agency serving law firms since 2012. He built Lexicon’s diagnostic approach to AI visibility around the same attorney credibility signals that answer engines actually check, and pushes every audit past surface fixes to the content and authorship gaps that keep firms unseen. David is a frequent contributor to Attorney at Law Magazine and Attorney at Work and a recurring guest on legal marketing podcasts.
