Volume 1 asked a simple question: can AI even find these local businesses? Mostly, yes. So this time we asked the question that actually matters, and the answer is the heart of this report. 
AI knows your business exists, and still hands the customer your competitors. Almost everyone is “findable.” Almost nobody gets recommended. That gap, not visibility, is the real story of this wave.
For the owners behind these businesses it’s starker still: AI can look up about 99% of them when handed their name, but names them as a recommended expert only about 9% of the time. Being findable turns out to be the easy part, and the part that doesn’t win you customers.
Then comes the twist that should stop any marketer cold: the things our whole industry optimizes for, Domain Rating, backlinks, organic traffic, barely predict whether AI recommends you. AI is playing a different game, and most businesses are still training for the old one.
(One honest note up front: raw scores jumped this wave, but mostly because AI now cites far more sources per answer, not because businesses got more visible. We treat that as a measurement quirk, not growth, and explain it plainly later.)
The Two-Volume Picture
Read the two volumes together and you can see the investigation evolving.
• Volume 1 (March) found that AI knew companies better than the owners behind them, and that nearly half of owners were effectively invisible. The question was visibility.
• Volume 2 (summer) found that visibility was never the point. AI can find almost everyone now. What’s scarce is being recommended, and that scarcity doesn’t track the SEO metrics you’d expect.
Each wave we dig deeper, and the picture keeps sharpening. Volume 1 asked whether AI could find you. Volume 2 found that being found isn’t enough, and uncovered more that business owners need to understand. We’re not done digging.
Volume 1 in Brief, for Readers Who Missed It
Volume 1, from a March 2026 reading, set the baseline. It asked whether AI could find these businesses and the owners behind them. What it found, verified and where needed corrected in an accompanying ledger:
• AI knew the company better than the owner. Companies averaged 37 out of 100; owners about 28. Nearly half of owners, 47%, were effectively invisible, though it’s worth noting Volume 1 scored this as one blended verdict; Volume 2 later separated being looked up from being recommended (see the owner section), so that 47% isn’t directly comparable to this wave’s owner numbers.
• Presence tracked with fortunes. Businesses AI recognized on both sides were growing traffic (a median of +3.0% year over year); those invisible on both sides were declining sharply (a median of -28.1%).
• Infrastructure wasn’t the problem. Almost everyone had the technical basics. What separated the visible from the invisible was earned, third-party coverage, which most didn’t have.
What We Corrected From Volume 1
Re-verifying every Volume 1 figure, we found and fixed several, all restated in the verification ledger, now the source of record rather than the original PDF. Two were genuine errors. First, the Visible tier (one of our four quality tiers) did not grow traffic as we’d reported; it actually declined about 14%. Second, the claim that mid-size firms had higher owner visibility than small ones ran backwards. A third figure moved for a cleaner reason: once we removed the misclassified businesses, the Both Visible quadrant, businesses AI recognized on both the owner and the company side, restated from +2.5% to +3.0% year-over-year traffic.
Note those are two different slices of the data that happened to share a starting number, so the -14% (a quality tier) and the +3.0% (a visibility quadrant) aren’t in conflict. We publish every correction plainly, the ones that hurt and the one that helped, because a study that tracks change over time is only trustworthy if it fixes its own record in public.
Why the Panel Is 372, Not 400
Volume 1 started with 400 businesses. On review, 28 didn’t belong and were cut, so the panel is now locked at 372 for every wave. The 28 were swept in by a keyword: 24 in Restoration, where “restoration” pulled in auto-body shops, museums, a botanic garden, a naval museum and nonprofits, none of them property-restoration companies, plus four in Home Services that were actually telecom, aviation and government. They’re kept in the workbook for audit; removing them just gives a cleaner sample, and it’s why Restoration is the smallest vertical (n=56).
The Recommendation Gap
This is the finding to build everything else around. When a real buyer opens an AI assistant, they rarely ask “tell me about Smith & Co.” They ask “who’s the best plumber near me?” So we asked both, for every business, and compared.
When asked about a business directly, AI could describe about 86% of companies. But when asked to recommend the best in the category, it named only about 1 in 5. That 67-point drop is the gap: businesses AI clearly knows, and still won’t put forward.
Being “known” by AI is nearly worthless. It’s the recommendation, the moment AI hands a ready-to-buy customer a name, that matters. And that moment goes to a small minority.
The Gap by Vertical
| VERTICAL | n | AI KNOWS THEM | AI RECOMMENDS THEM |
|---|---|---|---|
| Medical | 80 | 95% | 26% |
| Home Services | 76 | 89% | 22% |
| Restoration | 56 | 98% | 21% |
| Franchise | 80 | 88% | 14% |
| Legal | 80 | 65% | 14% |
Nobody’s winning. Medical does best and still gets recommended only about a quarter of the time. Legal is doubly stuck, AI knows only 65% of firms to begin with (it tends to know the attorney, not the firm), and recommends just 14%.
How we measured it: We checked whether AI named each business in its answer, so these figures are directional, not exact, read the recommendation figure as “about 1 in 5,” not a precise 19%. Two caveats worth naming. The “knows them” side is a low bar: our panel was built from a commercial database, so every business is already findable, and the check only asks whether AI gave a real answer when handed the name. The “recommends them” side is the clean test, there we never fed AI the name; we asked which businesses are best and checked whether each surfaced on its own. Either way, the 67-point gap between the two is far too large to be a matching quirk.
The Owner Opportunity
Volume 1’s headline was that the owner is the multiplier, that AI doesn’t just recommend companies, it recommends people, and that nearly half of owners were invisible to it. Volume 2 sharpens that into the single clearest opening in this whole study. 
A quick note if you’re comparing to Volume 1’s “47% of owners invisible”: that figure was a single blended verdict, part lookup, part recommendation, so it sits between these two numbers. It was never the recommendation rate. The clean recommendation figure is this wave’s 9%.
These owners aren’t hard to find. The “who should I trust for this” slot in AI’s answers is simply unclaimed. In a world where buyers increasingly ask AI who to trust, being the name it says pays off enormously, and almost nobody is that name yet.
Where the Slot Is Already Real, and Where It’s Wide Open
| VERTICAL | AI KNOWS THE OWNER | AI RECOMMENDS THE OWNER |
|---|---|---|
| Medical | 98% | 18% |
| Legal | 96% | 16% |
| Restoration | 100% | 4% |
| Home Services | 100% | 3% |
| Franchise | 100% | 1% |
In law and medicine, owner-as-expert already exists (16 to 18%), because those professionals built credentialed, third-party presence, bar directories, Healthgrades, real coverage. In the trades and franchises it’s 1 to 4%: almost entirely unclaimed, first-mover ground.
Why the Owner Number Swings, and Why That’s the Opening
Here’s the part that looks like a weakness and is actually the opportunity. Ask AI about a company twice, six hours apart with nothing changed, and the answer barely moves, about 0.4 points on our scale. Ask about the owner, and it can swing as much as 70 points; half the owners scored differently on the second identical run. Owner visibility is roughly 25 times more volatile than company visibility.
That instability isn’t random noise to fear. It’s the signature of an unclaimed reputation. AI swings because it’s guessing from thin signal. The owners who do get reliably recommended aren’t lucky.
They’ve given AI enough consistent, credible presence that it stops guessing. Volatility is what “nobody has built this yet” looks like in the data. Quiet the noise, and you become the stable answer.
The bar is on the floor. Volume 1 found the difference between invisible and recommended is often just two or three earned placements, a podcast appearance, a press mention, a contributed article in a trade publication. AI doesn’t need you to be famous. It needs you to exist in the sources it trusts. The first owners over that bar claim the slot before it’s contested.
Some honest limits worth stating plainly. “Can be looked up” means AI produced a confident answer to a “who is this person” query, not that we verified the answer was accurate, or even the right person; a common name or a confidently wrong bio would still count, so treat 99% as “AI will answer,” not “AI is correct.” Our panel also came from a commercial database, so every owner is already findable enough to be indexed, which means real-world recognition across all owners is lower.
And note the two waves measured this differently: Volume 1 scored owner visibility as a single blended verdict, while Volume 2 separates being looked up (99%) from being recommended (9%), so the raw owner rates across waves aren’t directly comparable, the clean numbers are Volume 2’s.
Because owner scores also swing sharply between identical searches, we report them only in aggregate, never as a single owner’s number. The opportunity is real precisely because the gap is, we’re not claiming AI favors owners or knows them accurately, we’re showing the expert slot is open and unclaimed, which is exactly when it’s easiest to take.
What AI Actually Decides to Cite
So why does AI know you but not recommend you? Look at what it’s reading. We pulled every source AI cited across the panel and sorted it by quality. The result is bleak, and it barely budged between waves. 
Even though AI now cites roughly six times as many sources as it did in March, the mix hardly changed, 81% weak in Wave 1, 78% in Wave 2. More citations, not better ones. And the newcomers to the top of the list are telling: LinkedIn now shows up for 99% of businesses, Yelp 91%, Facebook 69% and Reddit for 41%, a real sign AI is leaning on community and social content.
The Sources That Actually Carry Authority, by Vertical
| VERTICAL | WHERE REAL AUTHORITY COMES FROM |
|---|---|
| Legal | Super Lawyers, Avvo, Best Lawyers, Justia |
| Medical | Healthgrades, Zocdoc, Castle Connolly |
| Home Services & Restoration | BBB, Angi, HomeAdvisor, Houzz |
| Franchise | YouTube, Bizjournals, Entrepreneur |
The doors that count are a handful per vertical. Most businesses aren’t behind them, so even when AI knows you exist, it has nothing authoritative to recommend you on.
A caveat we take seriously: Wave 1’s stored sources were capped at ten per business, so exact “this domain rose from X% to Y%” deltas are directional. The quality-mix comparison (~80% weak in both waves) is the robust part.
Why SEO Doesn’t Buy the Recommendation
Here’s where it gets uncomfortable for our own industry. If real authority is what AI wants, you’d expect the businesses with strong SEO, high Domain Rating, lots of backlinks, real organic traffic, to be the ones AI recommends. We pulled that data from Ahrefs for all 372 and checked. They’re barely related.
Recommended vs. Not Recommended, on the Metrics SEO Optimizes (Medians)
| METRIC | AI RECOMMENDS | AI DOESN’T |
|---|---|---|
| Domain Rating | 22 | 15 |
| Organic Keywords | 44 | 33 |
| Referring Domains | 498 | 464 |
| Backlinks | 678 | 648 |
| Organi Traffic | ~0 | ~0 |
Read those as medians, the typical business in each group, not cut-offs. The 22 is the middle of the businesses AI recommends; the 15 is the middle of the ones it doesn’t. The two groups overlap heavily, plenty of businesses AI recommends sit below 15, and plenty it ignores sit well above 22, so there is no Domain Rating that “earns” a recommendation.
The story is how close the two numbers are. If link authority decided AI’s picks, the recommended group would tower over the other, say 45 versus 10. Instead it’s 22 versus 15, a nudge. That closeness is the finding: the metrics SEO optimizes barely separate the two groups.
A faint edge on Domain Rating and keywords; basically nothing on backlinks, referring domains, or traffic. Line these signals up against AI’s own authority scoring across all 372 and the correlations are weak to nonexistent (Domain Rating r = +0.10, referring domains +0.05, organic traffic 0.00). And it’s not hiding in one vertical, higher-DR businesses get recommended a little more in four of five, but in Legal it inverts entirely: the firms AI recommends actually have lower Domain Rating than the ones it skips, because legal recommendations run through Avvo and Super Lawyers, not your firm site’s link profile.
The scorecard our industry has optimized for two decades, Domain Rating, backlinks, traffic, barely predicts whether AI recommends a business. AI plays by different rules, and most businesses are optimizing for the wrong game.
This isn’t “SEO is dead”, the weak positive nudge says traditional authority still helps a little. It’s that the lever has moved. What gets you recommended is earned presence in the specific trusted sources AI actually pulls from, which is a different kind of authority than a backlink count.
One more reality check from the same data: about half these businesses have essentially no organic search traffic at all, and traffic level is identical for recommended and not-recommended businesses alike. Whatever drives AI’s picks, it isn’t how much Google traffic you already have.
Honest limits: The recommendation read is heuristic name-matching (noisy), AI’s authority score is itself a modeled number (so weak correlations partly reflect noise on both sides), and the Ahrefs traffic figure is a current snapshot, not a year-over-year trend. The consistency of the near-zero correlations is what makes us confident in the direction.
A Note on the Score Jump
You may notice raw visibility scores rose almost everywhere this wave. Don’t read that as businesses winning. It’s mostly a measurement quirk: AI roughly doubled how many sources it cites per answer, and since part of our score rewards being cited, everyone floated up together. When we strip that out and look at the quality of who cites you, the durable signal, company-level authority barely moved (about +1.5 out of 50, within the margin of error).
We kept the scoring identical to Volume 1 on purpose, and refused to “adjust” for the inflation, so the two waves stay comparable. We also ran a same-day reliability check: company scores are steady (they wobble ±0.4 when nothing changed), but owner scores swing wildly (±10, up to 71), so we report owner findings only in aggregate, never business by business.
The Five Verticals
Before the cards, here’s how to read them. Each shows three numbers. Company authority is the quality-of-sources score, and the arrow shows the change from Volume 1 (March) to now, so 13.5 → 13.9 means that vertical’s authority went from 13.5 in March to 13.9 today. We score it out of 50, not 100, on purpose. The full visibility score has two halves worth 50 each: how many sources cite you, and how good those sources are. The count half ballooned this wave when AI started citing far more sources per answer (the inflation we describe earlier), so it stopped reflecting real visibility. So we show only the authority half, the part that didn’t inflate.
AI knows and AI recommends are the share of the time AI describes, or recommends, that vertical. Every authority change below is small, which fits the panel-wide “held roughly flat” result; the real differences are in how often AI knows a business versus actually recommends it.
| Legal | n = 80 | |
|---|---|---|
| 13.5 → 13.9 Company Authority /50 | 65% AI Knows Them | 14% AI Recommends |
Legal is the toughest case. AI knows only 65% of firms, the lowest of any vertical, because it tends to know the individual attorney, not the firm. And it recommends just 14%.
It’s also where SEO helps least: the firms AI recommends actually have lower Domain Rating than the ones it skips. Legal recommendations flow through Avvo, Super Lawyers and Justia, not your firm site’s authority.
What it means: build the firm’s presence on the legal directories AI trusts, and build the attorneys as named experts. Your backlink profile isn’t the lever here.
| Medical | n = 80 | |
|---|---|---|
| 11.7 → 13.2 Company Authority /50 | 95% AI Knows Them | 26% AI Recommends |
Medical is the best of the bunch, and still only gets recommended about a quarter of the time. AI knows 95% of these practices, thanks to a dense directory world, Healthgrades, Zocdoc, Castle Connolly.
Company authority nudged up (+1.5) but stayed mid-pack. Being listed everywhere is table stakes; it isn’t the same as being recommended.
What it means: presence on the health directories AI leans on, plus genuine expert-verified coverage, is what moves you from the 95% who are known to the 26% who are picked.
| Restoration | n = 56 | |
|---|---|---|
| 12.6 → 14.3 Company Authority /50 | 98% AI Knows Them | 21% AI Recommends |
Restoration is the most visible, AI knows 98% of them, and recommends 21%. Its authority rose +1.6, but read that gently: it’s the smallest cohort (n=56) and a few members aren’t classic restoration firms.
What it means: in a vertical where everyone is ‘known,’ the BBB / Angi / HomeAdvisor presence that AI trusts is what separates the recommended fifth from the rest.
| Home Services | n = 76 | |
|---|---|---|
| 14.2 → 15.8 Company Authority /50 | 89% AI Knows Them | 22% AI Recommends |
Home Services carries the highest company authority in the panel and one of the larger moves (+1.7). AI knows 89% and recommends 22%.
The trades are cited heavily now, mostly by volume. Standing out still takes the trusted-directory presence (BBB, Angi, Houzz) that volume alone doesn’t buy.
What it means: being listed is a given; earning the review-site and directory authority AI reads is the path from known to recommended.
| Franchise | n = 80 | |
|---|---|---|
| 10.5 → 12.9 Company Authority /50 | 88% AI Knows Them | 14% AI Recommends |
Franchise had the biggest authority move (+2.4, still small) and the widest brand-vs-owner split. AI knows 88% of the brands and recommends 14%.
The lasting story is on the owner side: local franchise owners have the weakest personal presence of any vertical. AI knows the franchise; it’s fuzzy on who runs your location.
What it means: the brand-without-a-face problem persists. Building the local owner as a named, credibly-covered person is the opening.
What Changed Since Volume 1
• The question changed. Volume 1 asked if AI could find you. Volume 2 found that being found isn’t the win, being recommended is, and that’s rare.
• The SEO assumption broke. Traditional authority metrics barely predict AI recommendation. That’s new, and it’s the most important thing for a marketer to absorb.
• The surge is mostly plumbing. Scores rose because AI got chattier with citations, not because businesses improved. We caught it and set it aside.
• The scoring is provably identical across waves, recovered from the original code and validated. Two Volume 1 figures were corrected.
What This Means If You Run One of These Businesses
• Being findable isn’t the goal. AI already knows you. Getting recommended is a different, harder thing, and it’s where customers actually come from.
• Don’t assume your SEO carries over. A strong Domain Rating and backlink profile barely move whether AI recommends you. Useful for Google; not the lever here.
• Get into the rooms AI trusts. Recommendation flows from a handful of sources per vertical, the Avvos, Healthgrades, BBBs. Earning credible presence there is the real work.
• Don’t be fooled by a higher score. The whole field rose this wave on citation volume. A bigger number may mean nothing changed for you.
The game moved from “can AI find me?” to “will AI recommend me?”, and the answer runs through earned authority in the few sources AI trusts, not through the SEO scorecard most businesses are still chasing.
In Plain English: Two Games, and You Have to Play Both
If you read nothing else in this report, read this. Here’s what the data means for your business, without the marketing jargon.
1. Your Google Traffic Is Probably Slipping, But You’re Not Alone, and a Few Businesses Are Booming
We tracked the Google traffic of these 372 businesses month by month for a year and a half. The surprise isn’t that traffic is falling, it’s that the middle is disappearing. Almost half these businesses lost a big chunk of their traffic, more than a quarter of it, while about a third grew a lot. Very few stayed the same. The typical business is down somewhere between 7% and 14% over 18 months, but that number hides the real story: you’re either winning or losing, and fewer businesses are safely in between.
The total across everyone looks milder than that, but only because a handful of big winners mask how many are quietly slipping. One more sign worth noting: these businesses used to get a small traffic bump every summer, and this year it didn’t come.
2. Whether AI Recommends You Is a Completely Separate Question
Here’s the same five industries, side by side, on both games, what happened to their Google traffic, and how often AI actually recommends them.
| INDUSTRY | GOOGLE TRAFFIC (18 MO.) | AI RECOMMENDS THE BUSINESS | AI RECOMMENDS THE OWNER |
|---|---|---|---|
| Restoration | Grew (+17%) | Rarely (21%) | Almost never (4%) |
| Legal | Grew (+11%) | Rarely (14%) | Sometimes (16%) |
| Franchise | Slipping (-5%) | Rarely (14%) | Almost never (1%) |
| Home Services | Fell hard (-22%) | Rarely (22%) | Almost never (3%) |
| Medical | Fell hard (-31%) | Sometimes (26%) | Sometimes (18%) |
Read across your row and the point jumps out: doing well in one game tells you nothing about the other. Medical practices are recommended by AI more than anyone, and lost the most Google traffic. Legal and restoration grew their traffic, but AI rarely puts them forward. Franchises are slipping on both, and their owners are recommended just 1% of the time, the lowest of anyone. No industry is winning both games, which is another way of saying the field is wide open.
AI visibility and Google traffic are two different games now, and you have to play both. Winning with AI won’t save your Google traffic. Losing Google traffic doesn’t mean you’re invisible to AI. Most businesses are only playing one of them, or neither, and that’s the opening.
The traffic figures are the median business over 18 months, measured on the roughly half of each industry with usable Google-traffic data, so read them as direction and rough size, not precise numbers (per-industry counts run 21 to 42). The AI figures are how often each business or owner is named when someone asks AI who’s best, directional, not exact.
What We’re Still Digging Into
This is a living study, and some threads are still open, on purpose. We’d rather name them than paper over them.
• Traffic over time. We now have a full monthly Google-traffic series for the panel (Jan 2025 to Jul 2026), summarized in plain English above. The short version: the typical business is slipping, a minority are surging, the seasonal bump has faded and the trend doesn’t follow who AI recommends. We’ll keep this same monthly series running so every future wave has a clean, seasonally-smoothed comparison rather than a single volatile month.
• Owner-level reliability and accuracy. The owner questions are still too noisy to read business by business, and we haven’t yet verified whether AI’s confident owner answers are actually correct. Volume 3 will redesign the owner questions and add an accuracy audit, hand-checking a sample of AI’s owner descriptions against the real person.
• Reference anchors. We’re adding stable, well-known businesses measured every wave, so we can cleanly separate real change from shifts in AI’s own behavior.
How We Measured This, in Plain Language
We asked an AI system the same six questions about 372 businesses and their owners, twice, five months apart, keeping the questions and scoring identical so any change is real and not us moving the ruler. We cross-checked recommendation against what AI actually names in its answers, checked source quality against every domain it cited, and pulled independent SEO metrics from Ahrefs to test whether traditional authority explains AI’s picks. Where a measure is noisy, the owner scores, the name-matching, the snapshot traffic, we’ve said so and leaned on it only as far as it can bear.
Disclosures, in Brief
Recommendation and “known” rates come from name-matching AI’s responses, so they’re directional (read “~1 in 5,” not exactly 19%). “Known” for owners means AI answered a who-is query, not that the answer was verified accurate. AI’s authority score is a modeled figure; weak correlations with SEO metrics partly reflect noise on both sides. Ahrefs traffic is a full monthly series (Jan 2025 to Jul 2026); the typical business is down modestly over 18 months, results polarize into winners and losers, and traffic does not track AI recommendation.
This Ahrefs series replaces Volume 1’s single-month figures, which aren’t directly comparable, so the traffic series starts fresh at Volume 2. The panel-wide score jump is largely citation-volume inflation, not growth. Company scores are reliable within ~2 points; owner scores are not reliable per business and are reported only in aggregate; Volume 1 scored owner visibility as one blended verdict while Volume 2 separates lookup from recommendation, so the waves’ owner rates aren’t directly comparable. Scoring is identical to Volume 1, recovered from the original code and validated to reproduce every Volume 1 score.
All queries ran against Perplexity Sonar via API in both waves; findings describe that engine and were not cross-validated against other assistants. The panel is 372 after 28 out-of-category exclusions (Legal 80, Medical 80, Franchise 80, Home Services 76, Restoration 56); two Volume 1 figures were corrected in the verification ledger. Wave 1 citation lists were capped at ten per business, so domain-level deltas are directional while the authority-mix comparison is robust. A full methodology and disclosure document accompanies this report.
