You can pay for the past, or pay for the judgment.
The machine now does the activity. So the only thing left worth paying an operator for is deciding what the machine should aim at — and in AI search, that decision is the whole game.
A well-known analytics thinker put out a note this month arguing that marketers should renegotiate their agency contracts and expect to pay 25% to 75% less — starting now. His reasoning was simple and hard to argue with. AI got broadly capable, the ad platforms that sell you campaigns also own the models running underneath them, and every system now talks to every other system in real time. Work that used to justify a retainer is now done by the platform itself. His examples ran through paid media. But the math lands just as squarely on search, content, and AEO — and I think most people paying for those still haven’t done it.
I’ll say the uncomfortable part first, because I’m on the receiving end of it: a lot of what search agencies bill for every month is now busywork a machine handles for free. Manual rank tracking. Templated technical audits. Keyword lists generated on command. Weekly decks that re-narrate a dashboard the client already has open. If that’s what a retainer buys, the retainer should shrink. I’m not going to pretend otherwise to protect my own invoice.
But “pay less” is only half the sentence. The money shouldn’t leave the relationship — it should move to different work. And the work it should move to is the one thing the machine still can’t do: decide what to point it at.
Every sale today is judged twice. Once by the algorithm — Google, ChatGPT, Gemini, Perplexity — and once by the human on the other side. Winning both at once is judgment. It doesn’t come off a dashboard.
01 / The problem isn’t absence
Most businesses aren’t invisible in AI search. They’re mis-aimed.
When a founder tells me an AI model “never mentions us,” the reflex diagnosis is invisibility. It’s almost never that. The content exists. The pages rank. The business is real. The problem is that everything is pointing at the wrong target — the site answers a question no one is asking the model, uses a code or a category or a phrasing the model doesn’t associate with the buyer’s intent, and so the citation goes to a competitor who happened to aim straight.
Fixing that is not activity you can meter by the hour. It’s a judgment call about entities, architecture, and what a specific buyer actually types into a specific model on a specific day. A tool can generate a hundred pages. It cannot tell you which one deserves to be cited. That gap is what an operator is for.
02 / A contract that matches reality
Split the fee three ways.
Here’s the structure I now bring to any engagement where I can see real upside for the client. It borrows the shape of that analytics note and rebuilds it for search and AEO:
The base is the part I won’t zero out, and I’ll tell any prospect why. An operator who bets their entire fee on outcome isn’t showing confidence — they’re showing they need the deal. Real alignment is a floor that covers the work, plus a share of the upside I helped create. That’s skin in the game. Working for a lottery ticket is something else.
03 / The trigger is where honesty lives
Only tie a fee to something you can prove in a room a year later.
The outcome bucket is easy to write and easy to abuse. So I hold two rules. First, the trigger is a real, attributable outcome — net-new booked consultations from tracked organic and AI channels, or verified revenue through a booking flow I built and can attribute — never a self-reported or vanity number. Not rankings. Not “traffic.” Not a citation count on its own. If the platform reports it, the platform has every incentive to inflate it, and I won’t be paid on a number I can’t defend.
Second, the baseline goes in writing before month one, drawn from analytics I can see and stand behind. No baseline, no bonus, no argument later. And if a business is one where I genuinely can’t attribute the outcome cleanly — some are — then I don’t offer an incentive at all. I charge base and project, and I say so. Offering a bonus on a number nobody can measure is the same dishonesty as inventing a statistic. I don’t do either.
Analytics, search console, call tracking, whatever AI-citation tracking you run — it sits in your hands, not your agency’s. Without that you have no baseline, no proof, and no leverage to restructure. It’s the first move, not the last.
04 / What survives
The operators worth paying will be the ones selling judgment, not hours.
I don’t think this makes search work worthless. I think it makes the billing-for-rebuilds version of it worthless, and fast. The relationships that survive the next two years are the ones where the client pays a lean base for judgment and data ownership, project fees for the work that genuinely needs a strategist, and a modest, capped share of an outcome both sides can measure.
That’s the contract I want to sign, on the side of the table I’m sitting on. Not because it’s generous — because it’s the only version where the incentive and the work finally point the same direction. You can pay for the past. Or you can pay for the judgment that decides what the machine aims at next. I’d rather be paid for the second one, and I’d rather you only pay me if it lands.
◆
Sudhir Sharma
I’m an operator, not an advisor or an agency — I do senior search and AEO work, and I do it hands-on. I write the pages, spec the schema, run the outreach, and track the results. If there’s real upside in your search, I’ll take a share of it.