Over the past several weeks, I have found myself repeating a phrase that, at first glance, sounds dismissive.
“Then just don’t do it.”
The first time I used it, it was not intended as criticism, nor is it an attempt to diminish the challenges organizations face. Rather, it is an acknowledgment of a simple business reality. Every strategic initiative eventually reaches the point where execution requires meaningful investment. When an organization decides that the required effort outweighs the expected value, choosing not to proceed is a perfectly rational decision.
The problem arises when companies make that decision while continuing to expect the outcomes that only the investment would have produced; then it has morphed into a variation of the southern “bless your heart.”
I encountered this twice in the same week.
When Scale Becomes Sameness
At the International Search Summit, I presented research on canonical tokenization and semantic compression (when AI systems compress similar content into a single canonical version) and discussed how many multinational organizations are discovering that their international web strategies, while highly effective for traditional search, are increasingly problematic in AI-driven retrieval systems.
For years, global web programs optimized for scale and consistency. One English-language experience became ten, then twenty, and eventually forty nearly identical country sites. In many cases, the only meaningful differences were currency symbols, phone numbers, or shipping policies. Traditional search engines could distinguish these pages because technologies such as hreflang explicitly communicated that each version served a different geographic audience.
AI systems evaluate these pages differently.
When dozens of pages contain essentially identical information, many retrieval systems treat them as interchangeable evidence. Rather than presenting forty nearly identical sources, they often surface only the strongest or most authoritative market while suppressing the remainder. This is great if you are in a dominant market, but if you manage the web in Tanzania and you’re the 40th English version that was replaced, this is a big deal.
After the presentation, one attendee explained that is exactly the challenge they are having and that they had a consultant confirm that their hreflang implementation was correct. I agreed completely. Hreflang was designed to do exactly what it does for traditional search, but it has no place in AI systems.
The challenge is that AI systems are asking a different question. They are no longer asking whether two pages target different countries. They are asking whether those pages actually contain different knowledge.
The obvious path forward is to make each market genuinely local by incorporating regional terminology, examples, regulations, customer expectations, and buying behaviors that reflect the market itself rather than simply translating a master document.
The response was immediate.
“That would require an enormous amount of work.”
My answer was equally straightforward.
Then don’t do it.
Accept that those markets may continue to collapse into a single representative source within AI systems. That is a legitimate business decision. What is not reasonable is expecting a different outcome without changing the underlying information.
The Same Conversation in Product Data
Only a few days later, I found myself having an almost identical discussion with another enterprise organization, this time around product knowledge.
We were discussing that schema was not driving traffic as it had in the past. I had suggested expanding beyond the traditional Product Information Management (PIM) model of price, availability, image, and description to include more decision attributes. I believe that organizations should begin building structured knowledge that captures the decision process itself.
That includes decision attributes, compatibility matrices, feature translations between engineering terminology and customer language, problem-to-solution mappings, purchase eligibility, financing options, service availability, and other variables that help both customers and AI systems determine whether a product is appropriate for a particular situation.
I pointed to Google’s newer Merchant Center conversational attributes as the proof of a broader shift occurring in product information. Fields such as question_and_answer, variant_option, related_product, document_link, and similar attributes point to a future in which product data increasingly supports customer decisions rather than simply describing product specifications.
I suggested we collaborate with the teams to build out decision attributes, compatibility matrices, feature translations between engineering terminology and customer language, problem-to-solution mappings, purchase eligibility, financing options, service availability, and other variables that help both customers and AI systems determine whether a product is appropriate for a particular situation.
The first question was predictable. “Couldn’t AI generate all of that?” Perhaps portions of it could.
However, before discussing prompting strategies, it became clear that the more fundamental challenge was organizational. Someone would need to define those decision attributes. Different departments would need to agree on terminology. Product management, engineering, sales, customer support, and marketing would all need to contribute knowledge that had never previously existed in a unified structure.
The conversation shifted from AI to governance almost immediately. Eventually someone observed that this would require considerable effort across multiple teams. Again, my internal response was exactly the same.
Then don’t do it. But as an advisor, I opted to suggest starting small: let’s use AI and filters from the site, and build out slowly for the highest-value, decision-centric products.
The Competitive Advantage Is No Longer Knowing
Neither of these conversations was ultimately about AI but was about organizational willingness and change. Without digressing too much, this will be the driver of the failure of agentic search. We can have all the agents and bots we want, but if the information is not presented in an easy-to-consume manner, it is destined to fail, no matter how smart we make the ingestion logic.
Most companies already understand what they should do, and they do recognize that richer product knowledge creates better customer experiences; that is why they have salespeople and dozens of product information pages built into their website libraries. Too many are under the illusion that people are always ready to buy.
They understand that locally differentiated content is more valuable than globally duplicated content. If you have ever had to referee a market and HQ content session, you know this, but it does not fit in the scale and consistency box.
They appreciate that structured knowledge improves both search visibility and AI retrieval, and the limiting factor is rarely awareness of the need; it is whether the organization is prepared to invest the time required to build something competitors are unwilling to build.
This is where many digital transformation initiatives quietly stall. Not because the technology failed. Not because the strategy was incorrect. They stall because execution requires cross-functional cooperation, sustained governance, and knowledge engineering rather than another software platform.
AI Is Simply Exposing the Gap
One recurring theme in my recent writing has been that AI is not fundamentally changing what creates value. It is making existing weaknesses impossible to ignore. Organizations have always relied on undocumented knowledge scattered across product managers, support teams, sales representatives, and engineers. Human interactions compensated for incomplete documentation and fragmented information.
AI cannot compensate in the same way. It requires organizations to explicitly define the knowledge they previously assumed people would simply know.
That work cannot be delegated entirely to a prompt.
It requires the organization itself to decide what it knows, how that knowledge is structured, and how it should guide customer decisions.
A Decision Every Organization Must Make
There is nothing inherently wrong with deciding that the required investment exceeds the expected return. Every organization has finite resources and competing priorities.
What is increasingly unrealistic is expecting competitors willing to build richer knowledge structures, better decision models, and more complete representations of their products to compete on equal terms with organizations unwilling to make those investments.
The next generation of competitive advantage will not belong to the organizations with the best prompts. It will belong to the organizations willing to perform the difficult work of building knowledge that AI can confidently trust, retrieve, and explain.
If that work is not worth doing for your organization, that is a perfectly legitimate strategic choice.
Just recognize it for what it is.
It is a decision not to compete on that dimension.
