Firstly thanks to my friend Metehan Yeşilyurt from Peec AI for making me aware of their research (you should really read this report as I am just touching on one point or it here). One figure in Peec’s Q3 2026 report got me thinking. Of the companies that said they’d spent more on AI search over the previous year, three quarters said they couldn’t yet prove a return.
The survey covered 700 marketing decision-makers responsible for AI search at companies with at least 100 employees in the US, UK and Germany. It tells us what those people said about their spending. The more interesting question, at least for me, is what they should do when the next budget discussion starts.
Let’s assume you sell software. A potential client nearly rules you out because an AI says the CRM integration they need is available only on the enterprise plan. Your salesperson puts them right on a call. It’s included on the mid-tier plan.
How many other buyers got the same answer and never called? I’d want to know that too. But there’s a simpler question you can answer this afternoon. What does your own website say?
Start with the bit you can check
In this example, the pricing page says “CRM integrations” without telling you which plans include which integrations. An old support article still says enterprise only. Product confirms the policy changed six months ago.
OK, fix those two pages. Put the plan requirements where a buyer can find them and agree who’ll update the pages the next time the policy changes. You can do that without knowing where the LLM got its answer. A buyer reading the support article could make exactly the same mistake.
I’d still check the AI answer again and follow any links it gives me. When I followed Google’s AI recommendations for headphones under £100, one of the buying links led me to a different model from the one named in the answer. You’d miss that if you stopped after seeing the product name in the report.
When would I spend money on new content?
Let’s say you’ve corrected the pages and the answers about your company generally get the plan right. Then you ask which products work well with that CRM for a small sales team. Your software still rarely comes up.
Before commissioning a guide, ask sales whether prospects actually ask that question. Then read what the assistant cites when it recommends a competitor. Perhaps the competitor shows what data syncs, how long setup takes and what happens when the two systems disagree. Your page says the integration exists and leaves the buyer to work out the rest.
That’s a gap I’d consider spending money on. I’d give product marketing and SEO enough time to put together one decent guide, agree what it can cost and set a date to look at the results. Before publishing, record how often the product comes up and whether the answer gets the relevant details right. Use questions sales actually hears.
If there’s another integration with similar demand and a similar starting pattern, check that one over the same period. Make sure its basic product information is accurate before you start, then publish the new setup guide for just the integration you’re testing.
Why bother with the second integration? In a preprint by two researchers at Glasp, a web highlighting and video summary service, ChatGPT referrals grew 3.5 times on pages they’d left alone while they tested changes elsewhere on their site. They found some evidence that their changes helped, though it was hard to say by how much. Imagine how good the before-and-after chart would have looked if they’d only shown the pages they worked on.
If both integrations improve by about the same amount, I’d be careful about giving the guide credit. If the one with the new guide improves and the other stays broadly where it was, I’d look more closely. Did the answers start citing the guide? Did they get something right that they used to get wrong? If there’s little movement after a fair time period, you may have enough reason to stop. If the answers are all over the place or the platform changes halfway through, you may simply have an inconclusive result.
I made a similar point in my piece on rank tracking. A change in a report is a reason to look at what happened. How much you’re about to spend should have some bearing on how hard you look.
In this example, one potential customer happened to tell sales why they nearly ruled the product out. If somebody drops you from a shortlist before getting in touch, you may never hear why. That’s where an AI Search Audit can help. I look at the questions buyers ask while they’re comparing options, check what the answers say about you and follow the sources to see what makes competitors easier to recommend. That gives you a better idea of where it’s worth doing some work.
What can you actually measure about AI search ROI?
Lets pretend the guide starts getting cited and sales hears from two prospects who used it. One might have followed a link from an AI answer. The other might have found it on Google or received it from a salesperson. If the same guide helps people find you and gives sales a better answer to a question that keeps coming up, that strengthens the case for writing it. I’d still ask how buyers found it and what it helped them decide. That’s how you work out where to put the next bit of effort and how big this effort will be.
In Graphite’s n8n case study, the share credited to AI was 0.9% in GA4 last-touch reporting and 9% in a post-conversion survey. Those figures come from asking two different questions. Analytics gives credit according to the interactions it can see and the attribution model being used. The survey asks people how they heard about the product. Someone might get a recommendation from ChatGPT, search for the brand later and mention ChatGPT in the survey.
Graphite’s write-up gives us no survey sample size or response rate. I’d be comfortable saying the two methods gave very different answers. I’d be a lot less comfortable taking the 9% into a budget meeting and saying that’s what the AI search work delivered.
The IAB’s November 2026 attribution blueprint asks what role AI played in a purchase and what evidence would support a claim for credit. It also acknowledges how much of that evidence is missing or held by the platforms. For a content piece like this, I would start with what you can get hold of. Check referrals, ask sales what buyers said and see whether any of those conversations turned into qualified leads and opportunities. If a prospect tells you they were about to rule you out until they read the guide, write that down. It’s more useful than a number of everyone who has opened it.
Would you pay for a second batch?
Before writing another ten guides, I’d ask how many of your integrations have the same problem. You had a reason to write the first one. Buyers were asking about it, the product could do the job and the existing explanation was too thin. If that applies to two more integrations, you may have two more guides worth writing. Where did the other eight come from?
Then put a cost against the work, including staff time. Let’s say the next batch costs £6,000, one in four qualified opportunities becomes a customer, and each customer contributes £4,000 after variable costs over the period you’re looking at. On that assumption, you’d need six additional qualified opportunities to break even.
I’d underline additional in the budget meeting. At that close rate, you need six qualified opportunities you wouldn’t otherwise have had. If the guides mainly help sales close opportunities already in the pipeline, do that calculation separately. At £4,000 per customer, two extra wins would cover the £6,000. Either way, I’d ask sales if they’ve noticed that information from the guide may have changed a buyer’s decision.
Then ask what else those people could do with the time. Perhaps there’s a product page already bringing in qualified enquiries that does a poor job of explaining the offer. Perhaps the biggest problem is onboarding. If sales keeps hearing the same questions across several valuable product lines and the guides are helping buyers make a decision, a broader programme may be the better investment. That’s the sort of decision I help teams work through in strategic SEO consulting. Which opportunities are worth backing, what will it take to do the work properly, and what would give you the confidence to keep investing?
In The Pyrrhic Victory of GEO I put some thoughts together about the risk of moving money into AI search while neglecting the work it depends on. Here, that work started with an old support article and a pricing page. If the pilot gives you a good reason to write two more guides, fund two more guides. See what they do for buyers, then have the next budget conversation.
What would you fund next?
If you’re weighing up where to spend more on AI search, I can help you look at the questions buyers ask, see what’s getting in their way and decide what’s worth backing. We can work out what you’d need to see before investing more.

