Specific-Sounding, But Still Generic

The advice on specificity in writing is to add detail: name the tool, name the framework, add a number. So teams do. They swap "a CRM" for "Salesforce," swap "a lot" for "a 3x lift," swap "our approach" for "a data-driven methodology." The copy reads sharper. It is not more specific. It is the same generic paragraph wearing proper nouns, because every one of those nouns could sit in a competitor's paragraph without changing a word.

The short version: Adding brand names, tool names and round stats makes copy sound specific while it stays generic. None of it is indexical. None of it could only have come from your world. Real specificity is the presence of your world: the decision you made under pressure, the number your own operation produced, the thing you learned the hard way.

The machine can generate proper nouns endlessly. It cannot generate your particulars, because they were never in its training data. So name-dropping does not make you distinctive. It makes you a better-dressed version of the average.

Proper nouns are the cheapest detail there is

A language model reaches for named things the moment you ask it to "be specific," because named things are dense in its training data. It knows Salesforce, HubSpot, Slack, OKRs, North Star metrics. It hands them over on request. What it cannot hand over is the one decision you made at 2am that no dataset recorded.

That is the difference between sounding specific and being specific. A "40% increase in engagement" is a shape the model has produced ten thousand times; the number is round, the metric is standard, the sentence belongs to anyone. The reason it reads as evidence is exactly the reason it is not: it is the most probable continuation, the category average dressed as fact.

The test is indexicality, not detail

A detail is indexical when it points back to one source and one source only. "We tried three onboarding flows and killed the two that looked best on paper" is indexical: only your team ran that experiment, only your team learned that. "We use a proven, data-backed onboarding process" is not, even though it sounds concrete.

The tell is the swap. Take any sentence in your copy and try to put a competitor's name at the front. If it survives the swap unchanged, it carries no world. It was never about you. It described a category, and you pasted your logo on top of the category.

Why the sharpest-looking copy is the emptiest

There is a controlled version of this. In a study of writers using generative AI, individual pieces got better while the collective set grew more alike: each writer produced sharper, more polished work, and all of it converged. The polish is the convergence, because fluency is the one thing every model optimises for. Doshi and Hauser put it plainly: generative AI "enhances individual creativity but reduces the collective diversity of novel content."

Read that against your own drafts. The pieces that came out most fluent, most quotable, most stacked with names and stats are the pieces that read the way everyone else's reads. Fluency is not distinctiveness. Fluency is the thing the machine is best at, and the machine is best at the average. This is the same reason mentions and share of voice measure the wrong thing: the polished surface is not the distinguishing layer.

Switching to a bigger vocabulary will not save you

The instinct is to reach for rarer nouns: a more obscure framework, a more precise-sounding percentage, a sharper adjective. It does not work, because the pull is structural. RLHF narrows what a model returns toward the consensus middle, so prompting for "more specific" moves you to the average of everything already tagged specific. You get denser proper nouns and the same empty shape underneath.

The only detail a model cannot produce is the detail it never saw. Your customer's exact objection, in the words they used. The number your dashboard threw last Tuesday, the one that made no sense until it did. The reason you stopped doing the thing everyone told you to keep doing.

Specific-SOUNDING (generic underneath)	TRULY specific (indexical, first-party)
"We use Salesforce and HubSpot"	"We rebuilt our pipeline stages after three deals stalled at the same step"
"A 40% increase in engagement"	"The number our own dashboard threw when we cut the second CTA"
"A proven, data-driven methodology"	"The decision we made after the first version failed in week two"
"A best-practice onboarding flow"	"The onboarding flow we killed because our own users ignored it"
Round stats, tool names, jargon	A named decision only your operation could have made

Three tests for whether a sentence carries your world:

1. The swap test. Put a competitor's name at the front of the sentence. If it still reads true, the sentence describes the category, not you. Cut it or ground it.
2. The provenance test. Ask where the detail came from. If the answer is "everyone knows this" rather than "our operation produced this," it is borrowed, not owned.
3. The generation test. Ask whether a model can write the line without ever meeting you. If yes, it did not need your world, so it does not prove your world exists.

What being specific really costs

If a machine can generate every proper noun, then distinctiveness has to come from the one place the machine cannot reach: your particulars, supplied on purpose by someone who lived them. At Ivanooo we hold that in three layers, and none of them is decoration.

Voice is language a machine can measure as distant from the category center: not a rare word, a recognisable structure that survives the swap test. A model can average your voice; it cannot manufacture the decisions that shaped it. Entity is an identity the machine resolves by name, so your indexical detail attaches to you and not to the category you happened to mention. Topic Authority is the proof only you own: your data, your failures, your named decisions, the material no other brand's model could have produced because it was never in any training set.

Firoz Azees built the Ivanooo instrument to measure that distance from the average precisely because proper nouns cannot fake it. You can prompt your way to more names. You cannot prompt your way to a world you did not build. The brands that stay distinguishable stop decorating the average and start reporting from inside their own operation, sentence by sentence, because that is the one input a competitor's model does not have.

Run the check on your own page. Strip the brand names and the round stats, and ask whether anything is left that could only have come from you. If the page goes silent, you were sounding specific, not being it. See where your voice sits against your category's average: paste your URL, get the distance measured, with the evidence. No call required.

FAQ

Isn't adding specific details the standard advice for good writing? Adding detail is good advice; adding proper nouns is not the same thing. A tool name or a round stat is a detail a model can generate for anyone. Real specificity is indexical: it points back to one source, your operation, and survives no competitor pasting their name over it.

What makes a detail "indexical"? It could only have come from your world. A decision you made, a number your own dashboard produced, a failure you learned from. If a competitor can state the same line unchanged, it is categorical, not indexical, no matter how concrete it sounds.

Why does AI-assisted copy sound specific but read generic? Because a model returns the most probable continuation, and the most probable continuation is dense with the standard names and round numbers everyone uses. The polish is real; the distinctiveness is not. Individual quality rises while collective diversity falls.

Can't I just prompt the model to be more specific? No. Instructing a model to be more specific moves you toward the average of everything already tagged specific, so you get rarer nouns and the same empty shape. The one detail it cannot produce is the one it never saw: your particulars.

How do I test my own copy? Run the swap test. Put a competitor's name at the front of each sentence. Every line that survives unchanged describes the category, not you, and carries no proof your world exists. Rewrite those lines with a decision, a number, or a failure only your operation produced.