A Prompt to Make AI Text More Specific: Names, Numbers, and Real Examples

There is a pattern you will notice if you read enough AI output. The text sounds fluent, the sentences are grammatically clean, and the structure is logical. But when you finish reading, you cannot point to a single fact that stuck. No number. No name. No scene you could picture. Just a smooth wash of general claims.

This is not an accident. Language models default to the middle of the distribution. Specificity requires the model to commit, and committing risks being wrong, so without a direct instruction to get concrete, the model hedges toward statements that are technically true in all cases and useful in none.

The fix is a prompt. A specific one.

Why AI Writing Defaults to the General

When you ask a model to "write about" something, it optimizes for coherence and coverage. It wants to touch every relevant angle without saying anything that could be contradicted. The result reads like a summary of a topic rather than a take on it.

Real writing works the opposite way. A person who has actually done something says "the first three months, nothing grew" or "I paid $47 for a bag of perlite and used maybe a quarter of it." Those numbers are not decoration. They signal lived experience, and the reader trusts them in a way they cannot trust "results may take time" or "supplies can be costly."

The model is not incapable of specificity. It has seen enormous amounts of specific writing. It just needs you to require it.

The Three-Part Specificity Prompt

Here is a prompt structure you can drop into any existing draft or use at generation time. It asks for three distinct kinds of concreteness: quantities, named examples, and sensory or procedural detail.

Rewrite this section so it includes:
1. At least one specific number or measurable quantity (a timeframe, a count, a price, a percentage). If you do not have a real number, write a placeholder like [X weeks] so I can fill it in.
2. At least one named example: a real tool, a real book, a real person, a real place. If you are not certain the name is accurate, bracket it as [example: suggested name] so I can verify.
3. One detail that is sensory or procedural -- what something looks like, feels like, or what the actual step is, not just that a step exists.

Do not invent statistics. Do not fabricate citations. Use placeholders wherever you are uncertain, so I can supply real information.

The placeholder instruction is important. It keeps the model from hallucinating while still breaking the vague-default pattern. You end up with a draft that has the right shape for specific writing, with gaps you fill rather than fabrications you have to hunt down and remove.

How to Supply Your Own Numbers

The prompt above tells the model to use placeholders when it lacks real data. That placeholder system only works if you actually go back and fill them in. Here is how to do that efficiently.

Before you run the prompt, write a short "data block" at the top of your message. List any real numbers, names, or examples you want threaded into the copy. Something like this:

My data (use these, do not invent others):
- Timeframe: 6 weeks
- Tool I actually use: Notion
- Price I paid: $29/month
- Result I saw: 40% fewer revision rounds

Now rewrite the following section using the three-part specificity instruction above...

This approach means the model draws on real information you own rather than plausible-sounding data it generates. The final copy has specificity that is also accurate, which is the only kind worth having.

If you do not have your own numbers yet, the placeholder draft gives you a to-do list. Each bracket is a fact to verify or supply. That is a much faster editing workflow than reading a finished draft and realizing it is built on nothing checkable.

Combining the Specificity Prompt With a Persona Instruction

There is a difference between AI writing that contains specific details and AI writing that feels specific. The first is a matter of content. The second is a matter of voice.

When a model inserts a number into a sentence, the sentence can still read like a report. "Studies suggest a timeframe of six to eight weeks" is specific in one sense and lifeless in another. The specificity is attributed to a vague authority ("studies") rather than to a person with a reason to know.

To make specificity read like lived experience, pair the specificity prompt with a persona instruction that grounds the voice. Something like:

Write as someone who has actually done this. The specific numbers and examples should appear as things the writer noticed or measured, not as facts sourced from somewhere else. Use first person only where it fits naturally; otherwise let the specificity carry the credibility without announcing "I."

That persona instruction does not require you to write in first person throughout. It just asks the model to frame details as observation rather than citation. The result is the difference between "I measured six weeks" and "six weeks in, nothing had changed" -- the second is still grounded, but it does not lean on the "I" to do the trust-building work.

For a deeper look at how persona shapes everything from word choice to sentence rhythm, How to Tune a Humanizer Prompt to Your Own Voice covers that territory directly.

Where This Fits in a Longer Editing Workflow

The specificity prompt is one layer in a larger process. It handles the content problem (vague claims), but it does not address the pattern problems that make AI writing recognizable: the escalating adjectives, the three-item lists, the transitions that announce themselves.

A reasonable sequence looks like this:

1. Generate a first draft with a system prompt that sets tone and purpose. How to Write a System Prompt That Strips Out AI Tells has a template for that.
2. Run the three-part specificity prompt on any section that reads as a summary rather than a take.
3. Fill in the bracketed placeholders with real data you supply.
4. Do a final pass for pattern tells: the em-dashes, the words like "delve" and "crucial," the sentences that start with "It is worth noting."

The full humanizer prompt at /humanizer-prompt handles steps one and four together. The specificity prompt here handles step two. They are designed to work in sequence.

If you are deciding which AI tool to run this in, Humanizer Prompts for ChatGPT, Claude, and Gemini Compared breaks down how each model responds to specificity instructions so you know what to expect before you start.

And once you have specific copy, the next problem is usually that it still sounds neutral. How to Add Voice and Opinion to Flat AI Copy covers that next step.

Frequently Asked Questions

Can I use this prompt on any type of content, or is it only for blog posts?

It works on any text that suffers from vague generality: product descriptions, email copy, landing page sections, how-to guides, social captions. The three-part structure (quantity, named example, sensory or procedural detail) is format-agnostic. You may need to adjust what counts as a "named example" for each context. For a product description, a named example might be the specific material or a real customer situation. For a how-to guide, it might be a specific tool or brand.

What if the model keeps making up numbers even when I ask it to use placeholders?

Some models default to sounding authoritative and will fill brackets with plausible-looking data rather than leave them empty. If that happens, be more explicit: "If you are not certain this is accurate, write PLACEHOLDER in capital letters instead of a number." The all-caps signal is harder for the model to gloss over than a bracket. Then review every instance of "PLACEHOLDER" before publishing.

Does adding specificity help with AI detectors?

It can, but it is not a guarantee. Most AI detectors score on perplexity and burstiness patterns rather than content specificity alone. Specific writing tends to score better because it breaks the smooth probability distribution that detectors flag, but a draft full of specific facts can still score as AI-generated if the sentence rhythms are uniform. Treat specificity as a quality improvement first and a detector consideration second.

How many real data points do I need to supply before the copy feels credible?

One or two per section is usually enough. Specificity operates by implication: if a paragraph contains one genuine, verifiable detail, the reader extends more trust to the surrounding claims. You do not need to footnote every sentence. You need enough real anchors that the copy does not float.

What is the difference between this and just asking the model to "be more specific"?

"Be more specific" is a vague instruction, and models respond to it by adding adjectives or extending sentences. The three-part prompt forces a structural change: it requires the model to produce a number, a name, and a procedural detail, each of which is a different kind of specificity. The constraint is what makes the instruction work.