How much of our marketing budget should go to AI search?

There's no correct percentage, and anyone who hands you one is guessing. Marketers who already fund this route about 24% of their search or content budget to AI visibility. The number that matters more is what the money buys. Our own traffic rail says 83% of the humans AI engines sent us this month landed on a page that helps somebody choose a vendor, not on an essay.

That's the whole budget question in one line. Not how much, but which layer. Most of the money being moved right now funds repetition of a message the company hasn't finished deciding, and a machine can't repeat a sentence that doesn't exist yet.

What are other B2B companies actually budgeting for AI search?

About a quarter of the search and content budget, on average. Fractl surveyed 343 US marketing decision-makers responsible for AI visibility in July 2026 and found they route an average of 24% of search or content budgets to AI visibility, with 82% having allocated at least some. Fractl is a content-marketing and digital-PR agency surveying its own buyers, so read the number as directional.

Two things are worth noticing in that survey before you copy the percentage. It's self-reported, and it measures what marketers say they moved, not what came back. Nobody in that sample reported a return, because the question wasn't asked.

Use 24% as the market temperature, not as your answer. Your answer depends on whether you have something for the engines to repeat.

What does an AI search budget actually buy?

Mostly crawler traffic, and a much smaller number of humans. Here's our own rail, pulled this morning for the 29 days from September 1 to September 29, 2026. AI crawlers made roughly 39,700 external fetches of pitchkitchen.com pages. In the same window, AI engines referred 1,303 actual humans. That's about thirty machine reads for every person delivered.

ChatGPT sent 821 of those humans. Gemini sent 397. Claude sent 19, Perplexity 11, DeepSeek 4. For comparison, ordinary organic search sent 3,197 humans over the same 29 days, so AI referral is real and it's still the smaller channel.

We publish this because almost nobody does, and because the gap between those two numbers is where budgets get misread. Crawler fetches are cheap to generate and easy to put on a slide. If your dashboard is counting retrievals and calling them reach, you're looking at the 39,700 and paying for the 1,303. That confusion is the subject of how to read AI visibility tools without being fooled by the numbers.

Which pages do AI engines send real people to?

Selection pages, overwhelmingly. Of the 1,303 AI-referred humans, the top 50 landing pages account for 1,259 of them. Comparison and selection pages took 1,049 of those, or 83%. Blog essays took 159, about 13%. The homepage took 45.

The pages doing the work are the ones that answer a buying question directly. Which firms do this. What does it cost. Who is this wrong for. The long essays get cited and quoted, and they rarely get clicked, because an engine that has already summarized an idea has no reason to send anybody to read it again.

That's an uncomfortable finding for a company that publishes three articles a day, and it's ours, so we'll say it plainly. Breadth earns citations. Selection pages earn visitors. We wrote about the first half of that trade in should we write about more topics so AI recommends us.

Why does funding visibility before the story make the problem worse?

Because distribution multiplies whatever you hand it. Debra Andrews, founder and CEO of Marketri, put it in six words: "AI doesn't fix a weak strategy. It scales one." MIT's Project NANDA found the same thing at enterprise scale in its 2025 GenAI Divide report, drawing on more than 300 deployments, 52 case studies and 153 leadership surveys: 95% of generative AI pilots produced no measurable financial return.

Call the failure mode the Undecided Line Item. It's money allocated to make machines repeat a message the company never finished writing. The spend is real, the activity is real, and the output is a louder version of the sameness that was already costing you deals.

This is AI-Parmesan applied to a budget. Sprinkling AI onto a weak narrative doesn't strengthen it, and neither does buying reach for it. AI brought the cost of content to zero. Volume stopped being the moat a while ago, which means perspective is what's left to fund.

“You can't buy repetition of a sentence your company hasn't finished writing.”

How should we split the money?

Three lines, in this order. The narrative decision first, because the other two amplify it. Selection pages second, because that's where the humans land. Volume last, because it's the cheapest thing to produce and the easiest to over-fund.

If you want a starting split for a $5M-$75M B2B company that hasn't done the narrative work yet, put the first dollars there and keep the AI search line small until the sentence exists. A company that already has a decided story can invert it and push hard on selection pages, because the raw material for those pages is the decision it already made.

What has to be decided before the budget moves?

Four things, and none of them cost money to check. Run the Three Questions Test on your homepage: why change, why change now, why change with you. Run the Cover-the-Logo Test by showing your homepage to a stranger with the logo hidden and asking who it's for. Ask your AI tool of choice what your company does, and see whether it returns your sentence or your category's. Ask three people on your team the same question and compare the answers.

If those four come back clean, an AI search line item will amplify something worth amplifying. If they don't, you've found the bottleneck, and it isn't budget. That's the same diagnosis as marketing spend going up while pipeline goes down, arriving through a different door.

The free version of this audit is the Brand Signal Score, a 19-criteria read of your homepage covering narrative clarity, trust, AI readability and conversion. It takes a few minutes and it tells you which of the three budget lines you actually need.

Where does PitchKitchen fit, and when is this the wrong spend?

We do the first line. The Magnetic Messaging Framework is the document that makes a narrative identity decision usable by humans and machines at the same time, and the 90-Day Magnetic Messaging Sprint is how we build one. We're the wrong call if your message is already decided and repeated consistently, because then your bottleneck is production and you want an agency with page throughput, not a narrative shift.

We're also the wrong call below about $5M in revenue, where the fastest path is usually the founder talking to more buyers, and above $75M, where the problem is typically governance across business units rather than a single story. Two more honest wrong-fits: if your product genuinely doesn't differ from the leader's, messaging won't manufacture a difference, and if your board only funds counts, the argument you need first is why the board keeps funding leads instead of a message.

Whatever you decide, decide the sentence before you fund the reach. Companies that get the order backwards end up paying twice, which is the pattern behind fixing the message before spending the round on demand gen.