Is the AI work working?

Notes on fairness: July's position of 3.6 came from a short window right after your press weeks — a peak, not a resting level; 5.0 above the 5.5 baseline is the durable read. The branded-clicks row is deliberately kept red-side-up: it's the routing story (Question three), not a loss we're hiding. The two "not collected" rows are our own instrument gaps this cycle, named in the plan below.

The reading surged; the recommending hasn't caught up yet. Before an assistant recommends a product, its systems have to read and trust the material behind it. That reading broke out in July — a record week of 57 citations — and August is settling at about double the spring level. Assistant mentions sit at 5 in 100 conversations, unchanged within the margin since June.

Read it top-down. The citation surge follows the pattern we predicted: the May releases preceded the first stirring (Jun 8) by about a month, and the expert articles preceded the July surge by about two weeks. Honesty requires two caveats: your own site updates (mid-July onward) and peak lice season overlap the surge, so we say consistent with the work — not caused by it alone. And the bottom panel is the unfinished part: citations are the supply line, recommendations are the payoff, and the payoff line hasn't moved yet. In July we expected a 2–4 week lag; it's running longer. That's the number we're watching hardest this season.

July was a 7× month, and August is holding at roughly double the spring floor. A spike that vanishes is news; a spike that settles into a higher baseline is progress. So far August looks like the second kind — with the honest caveat that three weeks of volatile data (4, then 17, then 15 citations per week) isn't proof yet.

What stands out isn't just the volume — it's the breadth. The number of different pages being cited rose about 50% too, so this isn't one lucky article carrying everything; the expert-article library and the product pages are all being drawn on. Meanwhile your ordinary Bing search traffic barely moved — which tells us AI systems are picking up your content, not reacting to rankings. September answers the only remaining question: does the new floor hold?

The routing pattern held all summer — and the season makes it clearer. It's lice season: total search clicks, customer-question clicks and site visitors all rose. The one number that fell is clicks to your site from brand searches. Those searchers didn't disappear — you're shown more often than ever, at a better position than May — the results around you are Walgreens and Amazon, your own shelves.

We estimate roughly 4–11 brand-searches a week end at a retail listing instead of your site (the range depends on which earlier period you compare against). For a business that sells through 6,000 Walgreens doors and Amazon, that's intent arriving where you convert — but two honest footnotes. First, your position is holding around 5, better than the May baseline, though not the top-4 peak we reported in July (that was a short post-press window). Second, your own site updates this summer overlap these weeks, so we can't fully separate their effect. The fix we've recommended since July is unchanged and now seasonal-urgent: putting your real customer reviews into the product page's code, so your Google result shows stars the way the retail listings do — and the shelf-moment happens by design, not default.

One — and it's the first one that counts. On Aug 13, Google Analytics logged a visit in its "AI assistant" channel: someone clicked through to your site from inside an AI tool. One visit is a milestone, not a trend — but it's the first time the meter has moved, and it matches the citation surge upstream.

Direct visits nearly doubled this window (about 12 → 24 a day outside spike weeks) — but we won't claim that for the AI work. It's lice season, you know the brand is on shelves, and untagged links from your own emails and socials land in this bucket too. The mid-July peak is most likely the tail of your newsletter. One instrument note in the spirit of full transparency: the report that splits these visitors by landing page — our early-warning signal for AI referrals arriving unlabeled — wasn't collected this cycle, so that lens resumes next report. Remember: most people who see you in an AI answer never click at all — they search your name or walk into Walgreens. That's why the citation and brand-search numbers above are the real gauges.

We never say "caused." "Aligned" means the response arrived in the sequence and lag we predicted before it happened — twice now, on two different releases. The two open rows are where this report refuses to get ahead of the data.