Google's AI Overviews now strip 58 percent of the clicks from the top organic result (Ahrefs, May 2026), and your board is about to read that fall as decline. It usually is not. The engines answer the low-intent queries without a click, so the visitors you keep sit closer to buying.
But the fall in traffic is not the claim I am here to defend. This one is: in the AI recommendation set, legibility beats domain authority. When a buyer asks an AI to recommend a business like yours, the two or three it names are not the ones with the fattest link profile or the longest ranking history. They are the ones whose facts it can read cleanly. That is not where the inbound orthodoxy is looking.
The opponent here is the volume-first inbound model, the "more content, more traffic, more leads" playbook that HubSpot has spent years teaching, which still treats organic sessions and page count as the growth engine and would dispute every word of this. On that view a falling traffic line is a failure to be fixed with more pages, more keywords and more link-building. That playbook now loses money, because a traffic line that slopes gently down is the new default whatever you sell. The work is to be the source the engine can read and quote, and to measure what each remaining visit earns.
The visitors you keep are the ones closest to buying
Strip out the idle research and what survives to your site is the person who has decided to act. The "what is" and "how does it work" browsing gets absorbed into the answer. The visitor who still clicks through wants a price, a stock check, a phone number or a form.
That changes the arithmetic. If your sessions fall across a year while your qualified enquiries hold or rise, you have fewer visits but warmer ones, and often more real work to do. The traffic chart looks worse while the trade underneath it gets better.
The fair challenge is whether that pattern is really the mix improving, rather than seasonality, a price change or a new campaign flattering the numbers. You cannot prove causation from a single chart, and I will not pretend otherwise. What you can do is rule the alternatives out first: compare the same months year on year to strip seasonality, confirm the price and the enquiry definition did not move in the window, and check that no new campaign was running that could explain the lift on its own. With those held steady, what is left is the mix. If any of them moved, you attribute the change there first. The test that settles it is never the traffic line. It is the enquiries that line produces, and that is a question for the CFO.
The engines cannot name a brand they cannot read
Being named is now the whole game. The two or three businesses an answer engine lists get the consideration everyone else once fought for on page one. The rest are not ranked lower down, where a determined buyer might still scroll to them. They are absent from the sentence, and the buyer never sees the omission.
Being readable does two jobs at once. It lets a ready buyer act without friction, and it lets a machine lift your facts to cite you to the next buyer. The common failures block both: a price that lives only inside an image or a PDF, so it cannot be read, compared or quoted; opening hours that differ across two pages, so nobody trusts either; a contact form asking for eleven fields when the buyer had one question. Each one sends your most expensive visitor straight back to the answer engine.
The Crane Index™ does not poll the answer engines and count how often they name you. AI Overview and ChatGPT outputs are non-deterministic and personalised, so a name-count scraped from a handful of prompts proves nothing and cannot be reproduced next month. What the Index measures instead is the thing that decides whether you get named in the first place: how readable and well-optimised your own site is for a machine. It reads your live pages the way an engine does, server-side and deterministically, and scores three reads: whether an AI can retrieve and cite you, from your entity, your structured data and the facts it can lift as clean text; whether an agent can act on your site without hitting a wall; and whether a machine can confirm who you are from your own pages alone. The same site scores the same way every time it is read, which is what makes the number stable enough to compare month to month, and a lever you can pull rather than a weather report you can only watch.
The mechanism is blunt: a brand the engines cannot cleanly read is absent from the shortlist regardless of domain authority. A firm with strong authority and years of ranking history can still be missing from those recommendations if its fees and service list sit inside a designed PDF and an image banner, unreadable as text. Put the same facts on the page as labelled text, and it can enter a set that authority alone never bought it. Budget does not buy legibility.
The mechanism for the machine half is structured data, or schema markup: a standard vocabulary that labels your prices, locations, services and credentials so an engine can extract them cleanly. It guarantees nothing, and no one can promise a citation share, because there is no agreed standard for it. What schema does is remove the excuse to skip you. If the engine cannot extract your facts, it names a competitor whose facts it could.
The obvious rebuttal: if everyone labels their facts for machines, legibility stops being an advantage. Two answers. First, for most firms this is catch-up, and the window is open now, while the majority still hide prices in PDFs and contradict their own opening hours. The businesses that fix it this year gain the ground; the advantage is real precisely because so few have taken it. Second, legibility is not a one-time switch you flip and forget. Prices change, services change, and the way the engines read and reward pages keeps moving. The firms that keep their facts clean and current stay in the set, while the ones that labelled everything once and moved on quietly drift out. The moat is not the schema you added last year. It is the discipline of staying readable, and most firms do not sustain it.
What to put in front of a sceptical CFO
Stop reporting traffic as the headline. A CFO does not fund sessions. Report three figures instead, each read against your own number twelve months ago, not an industry average:
- Revenue per visit. Most of our clients take no payment on the site, so you do not read this off a checkout. You build it from the pipeline. Tag the session that produced each enquiry, follow that enquiry through to closed revenue in the CRM, and divide the settled revenue by the sessions in the same cohort. Because a lead-gen sale closes weeks or months after the click, fix an attribution window matched to your typical sales cycle and report each cohort only once it has matured, so you are dividing revenue that has actually landed by the visits that earned it, not this quarter's sessions against last quarter's deals. Rising means the mix improved even as volume fell. A word of honesty about small numbers: at these volumes the figure is noisy, and one large deal can swing it, so do not report it weekly on a handful of conversions. Wait until each matured window holds dozens of closed enquiries rather than a handful, and read it as a trend across several matured cohorts, not a single month. One month's revenue per visit on forty sessions tells you nothing. The same figure averaged across a year of matured cohorts is the one you take to the board.
- Qualified enquiries in absolute terms. Your early-warning line. If sessions fall while enquiries hold or rise, the mix improved and you can stop worrying about the chart. If enquiries fall in step with sessions, you have a real problem that no per-visit story will fix.
- How readable your site is to AI, measured by the Crane Index™. It reads your live pages the way a machine does and scores whether an engine can retrieve you, an agent can act on your site, and a machine can confirm who you are, the site-side conditions that decide whether you get named at all. Report the score and watch it move as you fix the failures above. Whether an engine names you on any given day is worth spot-checking by hand, but treat that as directional: those outputs shift with personalisation and have no agreed standard, so the number you take to the board is the readability the Index measures, not a naming count you cannot reproduce.
Get your measurement honest before you present any of this, because a number the CFO cannot trust is worse than no number. Do that, and the case writes itself: the volume-first shop still defending its traffic chart will spend two anxious years chasing volume that no longer describes its trade, while you take the work with fewer visits, warmer buyers and a name the engines can actually read.
