Before You Fund AI Search Prove It Can Reach a Buying Decision

AI search is already large enough to deserve attention. That does not mean it deserves your next pound.

My rule is simple: do not approve an AI search optimisation project until you can show that an answer system is touching a buying decision your business can legitimately help someone make. A screenshot of your brand missing from an AI Overview proves visibility was absent in one observed response. It does not prove lost demand, buyer relevance or a customer acquisition opportunity. That distinction sounds almost pedantic. It is actually the difference between a growth strategy and a novelty tax.

AI visibility is not the same thing as commercial opportunity

An AI answer becomes commercially relevant when three things overlap: the right buyer asks a consequential question, the answer shapes a decision, and your business has credible evidence that belongs in that decision.

Most AI SEO pitches begin somewhere easier. They collect prompts, count mentions and display competitors in brightly coloured dashboards. The theatre is effective because absence feels like loss. There is your rival. There is the answer. There is the empty space where your logo could have been. But an empty space is not yet a business case.

The prompt might sit miles from a purchase. The answer might summarise a topic without recommending a provider. The named competitor might be unsuitable for the buyer. Your company might have no defensible proposition to contribute. Even a useful mention might produce no visit because the reader has already received what they needed. I am not arguing that these observations are worthless. I am arguing that they are inputs. Treating them as the conclusion encourages marketing teams to fund the channel before defining the job.

Google itself describes AI Overviews and AI Mode as experiences that can surface links and help people explore complex questions. Its current guidance also says that established SEO fundamentals remain relevant and that there are no special technical requirements for appearing beyond normal Search eligibility. That should temper any proposal built around a mysterious new optimisation layer. Google’s guidance for generative AI features makes AI readiness look much more like disciplined digital publishing than a bag of secret prompts.

Start with the decision, not the prompt list

A useful opportunity assessment begins with a buying decision stated in plain English. Not “commercial solar queries.” Not “AI visibility for accountancy.” Not “own the conversation around rebranding.” Those are categories. A decision has consequences:

• Can this warehouse support the proposed solar installation without disrupting operations?
• Which accounting service fits a growing ecommerce company with cross-border VAT exposure?
• Is the company suffering from weak recognition, a confusing offer or a website that fails to convert existing demand?

These questions create different evidence requirements. They also reveal whether the opportunity belongs to AI search optimisation, conventional SEO, website design, branding, sales enablement, or no marketing intervention at all.

My standard is that the question must be specific enough for a competent salesperson or subject specialist to recognise what would change the answer. If nobody inside the business can describe the decision criteria, the content team is being asked to manufacture certainty the company does not possess. That is a poor foundation for AI marketing. It is an even worse foundation for trust.

The answer must have influence, not merely visibility

Some questions are informational cul-de-sacs. People ask, learn and leave. Others alter a shortlist, change a specification, expose a risk or send the buyer looking for proof. Both can generate impressions. Only one may justify meaningful investment from a customer acquisition budget. I separate answer presence from decision influence.

Answer presence asks whether an AI system responds, which sources it uses and which entities it names. Decision influence asks whether that response changes the buyer’s next action: what they compare, distrust, exclude, verify or investigate.

This is why a generic prompt such as “best web design agency” can be less useful than a narrower question about migrating a complex WordPress site without losing organic demand. The generic answer may contain more agencies and look more impressive in a report. The migration question exposes capability, risk, proof and fit. It creates a genuine decision surface.

That makes it less convenient than a visibility percentage, and more useful.

A mention can be true and still be commercially weak

Suppose an answer names your company. Good. Now ask why. Was the business included because a page contains a useful, specific explanation? Because an independent source corroborates a claim? Because a directory record happens to match the prompt? Or because the model produced a plausible name in an unstable response? Those situations should not be celebrated as one event called “AI visibility.” They create different risks and different next moves.

A brand mention unsupported by a clear buyer-facing proposition may create recognition without preference. A citation to a thin page may produce a visit that collapses on contact. A strong informational answer can earn attention while sending the wrong audience. A recommendation can be commercially valuable, reputationally dangerous or both if the stated fit is inaccurate.

This is where branding and conversion rate optimisation enter the argument. Being retrieved is only part of the journey. The public story must remain coherent when the buyer moves from an answer to a website, examines the offer and asks whether the evidence supports the promise. AI search can expose a weak brand faster. It cannot repair one by mentioning it more often.

Build the opportunity brief before commissioning the work

I would ask a marketing team to complete one page before approving an AI-search project. No 80-slide deck. No speculative traffic forecast dressed as inevitability. Four blocks are enough to reveal whether there is something worth pursuing.

1. The buyer-decision statement

Write one sentence: a ⟨defined buyer⟩ needs to decide ⟨specific choice⟩ under ⟨material conditions⟩. If the sentence cannot survive without words such as “solutions,” “innovative” or “best,” it is probably still a category statement. Tighten it until someone could disagree about the right answer.

Then name the commercial stage. Is the person discovering a problem, creating a shortlist, checking suitability, comparing risk or validating a final choice? The same topic can appear at several stages, but the content and evidence should not pretend those stages are interchangeable.

2. The observed answer behaviour

Record what actually happened across a bounded sample: the exact question and meaningful variants; platform, mode, location and date; whether an AI answer appeared; the decision criteria present or missing; sources and businesses named; changes across repeated observations; and the next action the answer appeared to encourage.

Do not turn ten prompts into a universal market claim. A sample is a sample. Its job is to expose patterns worth investigating, not to make the uncertainty disappear. Google now provides a generative AI performance report in Search Console for supported Search features, including page-level impression data. That first-party reporting can strengthen the observation layer where available, but impressions still do not answer whether the exposure changed a buying decision. Google’s Search Console documentation describes what the report contains; your commercial analysis must supply the interpretation.

3. The evidence entitlement

List the propositions your business is entitled to make. Then attach proof. That might include a documented process, product specification, named qualification, transparent pricing basis, original research, real project evidence or an expert explanation. Each item should help the buyer judge the decision, not merely make the business look busy.

This is the point at which many apparent AI-search opportunities turn into better projects. The real need may be to clarify the offer, publish missing technical evidence, reconcile inconsistent brand descriptions or improve the page that receives demand. All of that can support digital growth. None of it requires pretending that AI SEO is a separate species of magic. I would reject implementation if the proposed content needs to claim experience the business cannot document. Retrieval is not permission to invent authority.

4. The value and learning case

State what would count as useful progress and what the business expects to learn. Possible signals include qualified visits to a relevant page, assisted enquiries, sales conversations that reference a specific question, greater visibility for an evidence-rich resource or clearer source attribution around a decision. The correct measure depends on the journey. A high-consideration service should not be judged like an ecommerce product with a short purchase path.

Then write the stopping rule. For example: if repeated observation shows no relevant answer behaviour, no measurable exposure and no sales evidence after an agreed review period, the team pauses expansion and revisits the opportunity hypothesis. A stopping rule is healthy. It prevents an experimental channel from becoming a permanent retainer whose only proof is the dashboard it created.

Score readiness without pretending the score is truth

If the organisation needs a prioritisation tool, score each candidate decision from zero to two across five questions:

A low score does not prove AI search will never matter. It says the proposed work is not ready to compete for budget on the current evidence. A high score does not promise mentions, clicks or revenue. It means the opportunity is sufficiently defined to run a responsible experiment. That boundary matters. Marketing frameworks become dangerous when a prioritisation aid is promoted into a forecasting machine.

Sometimes the right AI-search project is a website project

The opportunity brief may reveal that the buyer’s question is real and the answer environment is active, but the company’s website cannot carry the decision. Perhaps the service page lists deliverables without explaining suitability. Perhaps a comparison article avoids the uncomfortable trade-off. Perhaps the About page obscures who is accountable. Perhaps strong evidence exists in sales documents but never reaches the public site. In those cases, the immediate task is not “get cited by AI.” It is to make the business easier to understand and verify.

That work sits at the intersection of website design, branding, content and growth marketing. Page hierarchy determines whether proof is discoverable. Language determines whether the proposition is precise. Interaction design determines whether the buyer can move from interest to evidence without fighting the interface. Sales knowledge determines whether the page answers a real objection rather than a keyword tool’s approximation of one.

This is my strongest objection to channel-first planning: it fragments one buyer experience into agency-shaped products. The buyer does not care which budget line paid for the sentence, interface or source. The buyer cares whether the whole explanation holds together.

What would make me fund the work

I would not require proof of guaranteed return; experiments exist because certainty is unavailable. But I would require a plausible chain between activity and customer acquisition. “Competitors appeared in screenshots” is not that chain.

There are also cases where strategic investment should begin before clean measurement is possible. A new search behaviour may matter to an industry with long sales cycles and sparse conversion data. A brand may reasonably build AI readiness because the same improvements strengthen conventional search, sales enablement and buyer trust. I can support that choice, provided it is described honestly as capability building, not disguised as a proven performance channel.

The language matters because it sets the expectation. Performance marketing asks for attributable movement. Capability building asks whether the organisation is becoming easier to retrieve, understand and evaluate. Both can serve business growth. Confusing them produces bad forecasts and worse relationships.

Take one decision into Monday’s meeting

Do not begin the next AI-search discussion with a platform tour. Bring one buying decision. Ask whether AI systems are visibly participating in it. Mark which criteria the answers cover. Put your public evidence beside those criteria. Inspect the website journey that follows. Decide what the team can learn within a bounded piece of work.

If the case survives those questions, fund the experiment with clear eyes. If it does not, do not fill the gap with more prompts and a larger dashboard. Fix the missing evidence, sharpen the brand proposition or spend the money where the buyer is already signalling intent. AI search deserves serious attention. It does not deserve exemption from commercial judgement.