Skip to content

Daily Blog

Follow the journey of building a GEO agency.

Day 140: Sometimes the Best AI Answer Is ‘It Depends’

A buyer asks, “Which supplier is best for our AI visibility programme?”

The question sounds complete. It is not.

Best for a company that needs a one-off diagnosis or a managed monitoring service? Best for a team with authority to change its website or one that needs board evidence first? Best for a single market or a multilingual launch? Best when the problem is category confusion, product misinformation, weak comparison language, or inaccessible source material?

A confident list of providers may be fluent while answering a decision the buyer has not actually defined.

For CMOs, Marketing Directors, and founders, the aim should not be to make every buyer question produce an immediate brand answer. Some questions should produce a branch. Some should expose a missing input. Some should end with a better follow-up question.

“It depends” is not a weak answer when the decision genuinely depends on facts the buyer has not supplied. It is the start of a more useful one.

Day 139: Stop Sending Product Gaps to the Content Team

A company is absent from an AI-generated shortlist for a commercially important buyer question. The familiar response is immediate:

We need a page for that.

Sometimes it does. The company may offer exactly what the buyer needs but explain it badly in public. A clearer page can correct a genuine communication problem.

Sometimes the missing page is doing the business a favour.

The product may not support the requested capability. The evidence may not exist yet. The offer may deliberately exclude that use case. Or the query may describe demand the company has no reason to pursue. Publishing around those gaps would not improve the truth available to a buyer. It would create a more persuasive version of the wrong answer.

This is where GEO becomes a question of decision rights rather than content volume. Marketing can communicate an approved commercial truth. It cannot approve a capability, redefine the offer, or choose a market on behalf of the functions that own those decisions.

Day 138: Your Help Centre Is Already Part of the Sales Conversation

A Marketing Director is evaluating a reporting platform before inviting suppliers into a formal process. She does not ask an AI system which product has the best dashboard. She asks a harder question:

If a scheduled board report fails on deadline day, who can restart it, how will we know what was delivered, and when do we need the vendor?

The campaign page promises automated reporting. The help centre explains the recovery path.

That operational answer may shape the buying conversation before sales knows the opportunity exists. It can reveal who needs administrative access, what the internal team must own, where vendor support begins, and whether the recovery route is acceptable for a high-consequence workflow.

This is a generalised buyer scenario, not a reported customer journey or an observation that every AI system will retrieve the same page. The commercial point is narrower: in AI-assisted B2B research, a serious pre-sale question may be answered by material written for existing users.

Your help centre is therefore not only a support destination. Parts of it may already be participating in evaluation. The response is not to turn documentation into sales copy. It is to preserve the practical answer and give the evaluator an honest route from operational fact to commercial fit.

Day 137: Test Whether an AI Agent Can Finish the Buyer’s Next Step

A procurement lead gives an AI agent a bounded task: compare two suppliers against an approved requirement, confirm whether one serves the UK, and prepare a consultation request for review.

The agent finds the supplier. It reads the offer accurately. It locates the contact page and fills in the visible fields.

Then the journey breaks.

The service selector is a styled element with no clear functional name. A required consent choice appears only after another field changes. The submit control becomes active without exposing why. After activation, the page shows a brief animation but no durable confirmation, reference, or clear statement that the request was received.

The supplier was visible. The content was understandable. The next commercial step was still unusable.

This is a generalised failure-mode scenario, not a reported buyer journey or a claim that AI agents routinely submit B2B enquiries today. It exposes a newer question for CMOs, Marketing Directors, and founders: if a person delegates part of a buying task, can the website represent each action and outcome clearly enough for the agent and the person supervising it?

Day 136: Answer Volatility Is a Citation Supply Chain Signal, Not an Algorithm Bug

A marketing team runs ten identical diagnostic prompts under controlled conditions to evaluate brand recommendations for a priority buyer question.

In four runs, the engine cites your brand as the primary recommendation. In three runs, it suggests a competitor. In two runs, it presents a generic category overview with no brand citations at all. In the final run, it surfaces a niche provider.

The common reaction inside marketing departments is frustration. Someone labels the AI erratic. Someone else calls it hallucination. A third person argues that tracking answer engines is pointless because outputs change across runs.

That reaction misses an operational signal.

While language models operate probabilistically, fluctuating outputs across identical diagnostic runs are not purely random noise to be ignored. When measured systematically under controlled run conditions, answer volatility functions as diagnostic telemetry.

It can signal that neither your brand nor your competitor holds unambiguous entity authority or complete evidence coverage across the retrieved sources for that specific prompt set.

Day 135: If AI Can Answer the Question, What Is Left to Buy?

A CMO asks an answer engine how to improve the company's visibility in AI-led research.

The answer is good. It explains that the company needs clear public information about its offer, accurate category language, useful evidence, accessible pages, and buyer questions worth investigating. It warns against treating one captured answer as market truth. It suggests looking at competitors, sources, and the routes a buyer may be offered.

Ten minutes later, the CMO understands the subject better. She still cannot responsibly decide what her company should change first.

She does not know whether the problem sits in positioning, product information, third-party evidence, technical access, or the questions being measured. She cannot tell which gap affects a live commercial priority. Nobody owns implementation.

That gap is where a credible paid offer begins.

As general explanations become easier to obtain, charging for access to the explanation becomes harder to defend. The commercial value has to survive the summary. It should sit in the diagnosis made under the buyer's conditions, the priority chosen among real trade-offs, the evidence created or verified, the change delivered, and the responsibility carried when the decision has consequences.

Day 134: A GEO Workflow Should Shrink the Decision Queue

Automation can make a GEO programme slower.

The team collects more answer-led observations, classifies more questions, refreshes more dashboards, and produces more frequent summaries. Leadership receives a larger stream of movement to inspect. Each fluctuation invites a meeting. Every absence looks like a task. The monitoring works, but the decision queue keeps growing.

That is not operational maturity. It is observation volume being converted into executive demand.

For CMOs, Marketing Directors, and founders, a mature Generative Engine Optimization workflow should do the opposite. It should absorb routine collection and first-pass classification, keep ordinary movement out of the way, and surface only a predefined exception tied to a commercial choice someone is authorised to make.

The valuable output is not another visibility report. It is a smaller queue of better-bounded decisions.

Day 133: Stop Ranking GEO Tactics Without Naming the Domain

Ask a GEO supplier which tactic works best and you may get a neat leaderboard: add statistics, quote experts, cite more sources, improve fluency, use authoritative language, introduce technical terms.

The list sounds actionable. It is also missing the decision that makes any item useful.

Useful for which content class? Helping which buyer do what? Supported by which source truth? Tested against which question and metric?

A statistic can clarify a research page and distort a service page when the number does not describe the service. Technical language can help a specialist and hinder an executive reader. Fluency can improve a page, but smoother prose does not repair an unsupported claim.

For CMOs, Marketing Directors, and founders, this is a budget-control problem. A universal checklist encourages teams to apply the same edits across unlike pages, then mistake activity for learning. The better question is not, “Which GEO tactic ranks first?” It is, “Which bounded intervention deserves a test on this content class for this buyer job?”

Day 132: Your Case Studies Can Teach AI the Wrong Customer Profile

A company can publish accurate case studies and still present a misleading picture of whom it serves.

The problem is not truth inside each story. It is the shape of the visible sample.

Perhaps the company built its reputation through large enterprise transformation programmes. Those projects generated recognisable logos, approved testimonials, and polished narratives. They became the public portfolio.

The business has since developed a lighter delivery model for mid-market teams. That offer is current, commercially important, and better suited to the buyers leadership wants next. Yet the public examples still show global rollouts, multi-quarter change programmes, complex procurement, and large implementation teams.

A suitable prospect may reasonably conclude that the company is too expensive, too complex, too enterprise-focused, or simply not designed for their situation. An answer-led system encountering the same accessible examples under recorded conditions may also frame the company around the segment that dominates the visible portfolio.

That does not mean the system has discovered the company's ideal customer profile. It means the public sample made one interpretation easier than others.

For CMOs, Marketing Directors, and founders, the distinction is commercially important: the customers you have served, the examples you can publish, and the buyers you want next are three different sets.

Day 131: Do Not Benchmark GEO Only Where Competitors Already Win

A competitive benchmark can make your GEO strategy less competitive.

It happens when the research starts with a rival's public footprint. The team collects the categories the rival names, the comparisons it appears in, the questions where it is already prominent, and the criteria its pages explain best. Those inputs become the benchmark. The resulting report shows where your company trails.

Every finding may be accurate within the recorded checks. The strategic frame can still be wrong.

The benchmark has allowed the competitor to choose the questions, define the vocabulary, and set the boundaries of the contest. Your company is measured inside territory the rival has already shaped. The apparent response is to reproduce the same comparisons, answer the same questions, and compete on the same criteria.

For CMOs, Marketing Directors, and founders, this is not merely a measurement problem. It is a portfolio-allocation problem. A useful benchmark should show where parity matters, where a rival deserves a direct challenge, where buyers face an unresolved decision, and where the company should decline to spend at all.

The goal is not a longer prompt list. It is to stop competitive measurement becoming imitation.