The Prompt That Broke Me: Why AI Can’t Preserve Meaning

Semantic drift is becoming one of the biggest hidden risks in AI-generated content.

I used to love AI for writing.

It was fast, creative, and made me feel like a productivity god.

Then it started hallucinating my life.

I was updating my CV. Simple task, right?

I fed in my experience, achievements, and a clear brief:

“Professional tone. Keep it factual. Emphasise specific skills. No embellishments.”

The first output was decent.

The second time I ran the prompt, it drifted.

Suddenly I was “a visionary leader transforming industries” instead of the guy who delivered products on time and under budget.

By the third iteration, it had rewritten my career narrative into something I barely recognised.

The words were polished. The structure looked better. But the semantic meaning, the core truth of who I am and what I’ve actually done, had quietly slipped away.

That was the moment I realised something important:

AI is incredibly good at generating language. It is not naturally good at preserving meaning.

Semantic drift compounds across prompts, iterations, teams, systems, and publishing channels.

So I did what most people do. I added more prompt instructions.

Then more.

“Do not add flair.”

“Preserve original meaning.”

“Stick exactly to these bullet points.”

“Never use these terms.”

“Maintain this positioning.”

Eventually, the prompts became longer than the actual task.

And it still drifted.

At some point I realised I was spending more time correcting the AI than I would have spent writing the content myself.

The AI was improving the writing while quietly degrading the truth.

Worse, this wasn’t just happening with CVs.

It was happening with:

• blog posts
• proposals
• social content
• product positioning
• strategy documents
• client communications

Anything important enough that meaning actually mattered.

That frustration became the starting point for Publish Weaver.

At first, it was just a simple internal application. A way to lock semantic meaning in place and stop AI systems wandering away from authoritative intent.

Nothing fancy.

• Canonical terminology
• Semantic constraints
• Validation rules
• Approved positioning
• Source traceability

But the deeper I went, the bigger the problem became.

This wasn’t really a prompt engineering issue.

It was a governance problem.

AI governance today focuses heavily on models, security, compliance, lineage, lifecycle management, and risk scoring.

But almost nobody governs semantic integrity: whether published content still means what it was originally intended to mean.

And that gap becomes catastrophic at scale.

Marketing rewrites messaging.

Sales paraphrases it.

Regional teams localise it.

AI reinterprets it.

Partners summarise it.

Procurement modifies it.

External systems regenerate it.

Over time, organisations lose control of their own meaning.

That is when Publish Weaver stopped being a tool and became a governance system for external-facing content.

Not just a prompt wrapper.

Not just workflow software.

Not just AI guardrails.

A governance system designed to provide:

• semantic governance
• publication assurance
• provenance
• validation
• operational governance
• auditability

Because the real risk with AI is not simply hallucination.

It is semantic drift at scale.

A slight deviation repeated thousands of times across AI-generated content, distributed teams, automated workflows, and external publishing surfaces.

One small drift becomes organisational confusion.

Then compliance risk.

Then reputational risk.

Then trust erosion.

Publish Weaver exists to stop that drift before publication happens.

Not by slowing AI down, but by making AI-generated content governable.

We spent the last few years teaching machines how to generate language.

The next challenge is ensuring generated language remains aligned, authoritative, and trustworthy.

Because once organisations lose control of meaning, they eventually lose control of trust.

That is the problem Publish Weaver was built to solve.

Read more

Publish Weaver Blog Article