What Happens When an AI Assistant Repeats Your Abandoned Positioning?

When an AI assistant repeats your abandoned positioning, treat it as evidence that your public record still supports the old story. Audit the sources behind the answer before rewriting the website, then make every durable public reference agree with the position you want a partner to evaluate.

In 1985, Coca-Cola introduced New Coke in the United States after testing a sweeter formula. The company had enough research to believe the change could work. By July, after intense public resistance and uncertainty about what the reversal would mean, it announced the return of the original formula as Coca-Cola Classic.

Mark Pendergrast documents the episode in For God, Country and Coca-Cola. The failure was not a lack of information. Coca-Cola had information that answered one question, taste preference, while missing another: what the existing product represented in people’s lives.

That distinction sat in front of me on a Friday, while I was preparing for a partner call.

I asked an AI assistant to describe the company. It gave back language we had stopped using. It described a broad service identity that had made sense when we were taking almost every reasonable piece of work we could find. The present work was narrower: practical AI and automation decisions, built around products that had to survive real constraints across Africa, Germany, and the US.

The answer was polished enough to be plausible. That was the problem.

A clean homepage can lose to older evidence

The website had changed. Old pages, directory profiles, launch announcements, social bios, interviews, and scattered references had not all changed with it.

A partner who lands on the site sees the current claim. An AI assistant may see the accumulated record, including pages that were once accurate and are now misleading. If it cites an old announcement, the problem is clear. If it gives no citations, the answer is still a useful warning: the old positioning remains easy to retrieve.

Friday gave me a choice that looked smaller than it was. I could rewrite the homepage again, making the new position louder. Or I could accept that the homepage was already saying the right thing and trace the wider evidence trail that kept pulling the answer backwards.

Rewriting the site would have felt productive. It also risked treating the visible page as the source of the problem.

The evidence trail decides what people repeat

Positioning is partly what you say today. It is also what other people, platforms, and models can still find when they look for a shortcut to understanding you.

For a founder with limited runway, this matters before a partner call, a hiring conversation, or an automation engagement. Someone evaluating you may spend ten minutes across your site, search results, a profile page, and an AI answer. They are building confidence from consistency. Conflicting descriptions create work for them, and work creates hesitation.

I started with the exact words the assistant used. Then I looked for their likely origins:

• Old service descriptions that still ranked or remained indexed.
• Profile pages that described a previous version of the business.
• Articles whose framing reflected an earlier market and earlier constraints.
• Bios written when a broad claim felt safer than a precise one.

This is less glamorous than a new positioning workshop. It is closer to product maintenance. A broken state in one part of a system can become the state a user experiences, even when the main screen looks correct.

The useful question was not, “How do we make the assistant say the right thing?” I cannot control that answer directly. The useful question was, “What public facts would make the old answer harder to defend?”

Update claims where they can keep teaching the market

The weekend task became a short list of high-value corrections. The homepage needed one check for ambiguity. Then the work moved outward: the pages likely to appear in search, the profiles a partner might open, and the old descriptions that could still be quoted or reused.

Some material deserves to stay because it records a real stage of the company. Historical work should remain historical. The problem begins when an old statement is presented as current.

That means adding dates where a past offer or market focus needs context. It means changing a bio that has outlived the work it describes. It means removing a claim that sounds impressive but leaves a buyer unsure what decision you can actually help them make.

This is the same discipline behind AI Code Approval: What Daniel Learned About Ownership and Human Review. The useful output is not the fastest automated answer. It is a process where someone can inspect the decision and take responsibility for it.

For positioning, the owner needs to be able to explain why each public claim is still true.

Let the partner call expose the remaining gap

By Monday, I did not expect every old reference to disappear. Public information has a long memory. I wanted a defensible current record and a clear way to handle the gap if it surfaced in conversation.

That preparation changed the call. Instead of hoping the other person had formed the right impression, I could state the current focus plainly, name the earlier work as earlier work, and explain the decision that connected them.

Coca-Cola could not make people forget New Coke by repeating a slogan. It had to act on the evidence that its public understanding of the product had diverged from its own model. The same is true when an assistant revives a position you left behind.

Run the test before the call. Save the answer. Inspect the sources where you can. Then fix the public claims that are still teaching the market to describe you incorrectly.