A tactic can be legitimate without being significant.

As brands rush to adapt to AI search, a new category of digital marketing has emerged: things companies can point to as evidence that they are “AI-ready.”

An llms.txt file.

An AI information page.

A chatbot.

A new block of schema.

A press release announcing an AI strategy.

Instructions telling AI systems how the company should be described.

None of these things are inherently bad.

Some are genuinely useful.

The problem starts when having the tactic becomes confused with accomplishing the objective.

That is where AI readiness becomes AI theatrics.

Apollo provides a useful example

Apollo.io recently published a substantial /llm-info page specifically for AI assistants. It contains company facts, products, positioning, integrations, executives and other information an AI system might need. It even includes explicit instructions for AI assistants about how Apollo should be characterized.

That’s interesting.

It is also perfectly reasonable.

If an AI system retrieves the page, Apollo has given it a clean, current, first-party source instead of forcing the system to reconstruct the company from hundreds of pages scattered across the web.

But the existence of that page tells us almost nothing about whether Apollo is actually winning in AI discovery.

It does not prove that ChatGPT retrieves it.

It does not prove that Claude trusts it.

It does not prove that Perplexity cites it.

It does not prove Apollo appears when buyers ask which sales platform they should consider.

And it certainly does not mean an AI system will obediently follow Apollo’s preferred description simply because Apollo wrote instructions telling it to.

Publishing information for AI and influencing what AI actually knows, retrieves, trusts and recommends are very different things.

Apollo itself demonstrates the distinction.

Beyond its /llm-info page, Apollo also offers an MCP server, API and other integrations that allow AI systems to access Apollo data and perform actual sales workflows. That moves beyond describing Apollo to an AI system and toward allowing an AI system to use Apollo.

One is information packaging.

The other is utility.

Neither automatically guarantees business impact.

Four very different levels of “AI-ready”

Brands should separate four layers that are increasingly being lumped together.

Layer	The real question	Examples
AI signaling	Can we point to something AI-related?	AI page, chatbot, llms.txt, “AI-powered” messaging
AI legibility	Can AI systems reliably understand our business?	Crawlable content, structured data, consistent entities, clear product information
AI utility	Can an AI system actually do something useful with us?	API, MCP server, connector, product feed, transactional integration
Business impact	Is any of this changing outcomes?	More citations, recommendations, qualified traffic, leads, transactions or efficiency

There is nothing wrong with Layer 1.

It simply should not be confused with Layer 4.

The new vanity metrics

We’ve been through this before.

Twenty years ago, brands wanted to know how many website hits they received.

Later it became social followers.

Then keyword rankings.

Then impressions.

Every technology shift creates metrics that are easy to measure, easy to report and only loosely connected to the thing the business actually cares about.

AI is producing its own versions:

“We added llms.txt.”

Useful, perhaps. The current llms.txt proposal is designed to make information easier for agents to consume, and adoption has grown considerably. But the specification itself is about providing cleaner information to AI systems. It does not promise visibility, citations or recommendations.

“GPTBot crawled our website.”

Interesting, but even that needs context. OpenAI distinguishes GPTBot, which is associated with potential model training, from OAI-SearchBot, which is used for ChatGPT search discovery. A bot visit is not the same thing as appearing in an answer.

“We created an AI information page.”

Possibly valuable. But only if systems find it, use it, reconcile it with other information and ultimately change an answer because of it.

“We added schema.”

Good structured data can reduce ambiguity. Bad, unsupported or irrelevant structured data can simply provide machines with a cleaner representation of bad information.

“We launched an AI chatbot.”

That may improve the customer experience. It says almost nothing about whether external AI systems understand or recommend the company.

The common mistake is measuring deployment instead of effect.

What actually matters

AI systems do not encounter brands through one magic file.

They encounter an information environment.

Your website matters.

Your product pages matter.

Structured data matters.

Documentation matters.

Customers matter.

Reviews matter.

Industry publications matter.

Partners matter.

Reddit discussions matter.

Third-party databases matter.

APIs and feeds can matter.

How frequently information changes matters.

Whether different sources agree with each other matters.

And ultimately, the question is not whether your brand completed an AI checklist.

It is whether the information available to AI systems makes your company easier to understand, retrieve, validate and use.

A better AI-readiness checklist

There are still plenty of worthwhile quick wins.

Create one authoritative, human-readable page explaining the company, products, customers and differentiators.

Make important information accessible to crawlers rather than burying it inside inaccessible interfaces. OpenAI specifically recommends allowing OAI-SearchBot when publishers want content discoverable in ChatGPT search.

Use structured data to reinforce visible facts rather than trying to manufacture them.

Answer the questions buyers actually ask instead of producing generic “AI optimized” content.

Keep product, pricing, location and company information consistent across the web.

Build third-party authority through customers, partners, reviews, publications and communities.

Use llms.txt, AI information pages and machine-friendly documentation where they genuinely make information easier to retrieve.

Build an API, MCP server or connector when there is an actual workflow an AI system could perform.

And assign someone responsibility for keeping all of it accurate.

Those are useful actions.

But none should be considered successful simply because they were deployed.

Measure the answers, not the artifacts

A much better scorecard starts with questions like:

When someone asks about our category, are we included?

When someone asks for recommendations, are we considered?

Are our products and capabilities described accurately?

Which competitors appear when we don’t?

What sources are AI systems relying on?

Are incorrect statements disappearing as we improve the underlying information environment?

Are citations, recommendations or qualified visits increasing?

Can an AI assistant successfully complete a useful action involving our company?

And ultimately:

Is any of this creating business value?

That is a much higher standard than checking whether /llms.txt exists.

It is also considerably harder.

The distinction brands need to make

The AI transition is real.

Consumer discovery is changing.

Interfaces are changing.

The companies that make themselves understandable and useful within those systems will have an advantage over companies that ignore the shift.

That makes it even more important not to waste the opportunity on theatrics.

An AI page can be useful.

Schema can be useful.

llms.txt can be useful.

A chatbot can be useful.

An MCP server can be extremely useful.

But they are inputs, not outcomes.

If an AI tactic does not improve what AI systems know, what they can do, what they cite, what they recommend or what users accomplish, it may be signaling innovation more than creating it.

The goal isn’t to look AI-ready.

The goal is to become AI-useful.