Table of Contents
TL;DR
- AI tools default to generic copy because the prompt did not name the buyer, the position, or the voice. The tool fills the vacuum with the statistical middle of its training data, which reads as no voice at all.
- Three failure modes produce almost every generic page. Missing positioning — the brand has not decided what context the buyer should place it in. Missing voice — the brand has not decided how it sounds. Missing customer language — the brand has not written down the words real buyers use.
- Positioning is the upstream fix. April Dunford’s five components anchor the work: competitive alternatives, unique attributes, value with proof, best-fit buyer, market category.
- Voice is the second fix. A short spec — five sentences on how the brand sounds, three on how it does not, three banned phrases — gives the writer and the tool a target.
- Customer language is the third. The vocabulary lives in the support inbox, the reviews, the recorded sales calls. The same edit that fixes generic copy also earns AI citation lift on chat surfaces.
A homepage hero reads cleanly. Three benefits, a CTA, a logo bar, a testimonial. The page looks professional. The page also looks like every other homepage the buyer scrolled past on the way here.
The AI tool produced the draft in 60 seconds. The buyer left the tab in five.
A generic page is a hotel-lobby painting. Pleasant, professional, and impossible to remember.
The fix is not a punchier headline or a longer hero section. The fix sits upstream. The page reads generic because three decisions were never made before the prompt was written.
Naming those decisions and feeding them to the tool changes what comes back. The same edit also gives answer engines specific phrasing to cite when they summarize the page on a chat surface.
Why does AI website copy read generic by default?
An AI tool has read the entire internet. It remembers no client specifically.
Whatever the prompt leaves out, the tool fills with the statistical middle of its training data. The middle of the training data is, by construction, average prose. The output reads as the average page in the category because the prompt described the category, not the company.
A prompt that names the buyer, the position, and the voice produces copy that can only belong to one business. A prompt that names a category produces copy that reads like every other page in the category. The shape of the output mirrors the shape of the input.
This is the shift most writers miss. In the agency era, the project notes went to a human who filled in the context gap with judgment and experience. Half-finished project notes still produced usable copy because the writer carried the missing half in their head. The AI tool carries none of that.
The decisions that close the gap belong upstream of the prompt, not inside it.
What does "generic copy" actually mean on a website?
Generic copy is a surface symptom of three deeper problems.
The page reads as a description of a category rather than a position inside one. "We make project management software" is a category sentence. "We are the project tool the founders of professional services firms use to stop being the bottleneck on every client report" is a position.
The page reads in the median voice of the training data. The buyer cannot tell whether the brand is warm or formal, plain or technical, confident or hedged. The page sounds like every other page because no register was committed to.
The page contains adjectives and abstract nouns where the buyer’s actual words would belong. "Streamlined." "Efficient." "Best-in-class." None of these words appear in the buyer’s vocabulary, so none of them register when the buyer reads them.
The buyer reads two paragraphs and senses they could swap to a competitor’s tab without losing information. The page never earned the second click. Generic is not a style problem. Generic is a missing-decisions problem.
Which three failure modes produce generic copy?
Three failure modes account for almost every generic page on the open web.
Missing positioning. The brand has not decided what context the buyer should place it in. The hero reads as a feature list rather than an argument.
Missing voice. The brand has not decided how it sounds when it talks. Every page reads in the same flat register because no register was specified.
Missing customer language. The brand has never written down the words real buyers use. The page is full of marketing-team vocabulary that the buyer does not recognize.
Each failure mode has a separate fix. Most pages need work on more than one. None of the three is solved by a better adjective or a longer page.
The diagnosis comes first. The rewrite comes second.
A page without positioning is a tactics problem misdiagnosed. A page without voice is a brand problem misdiagnosed. A page without customer language is a research problem misdiagnosed. The fix that works depends on which diagnosis is correct.
How does positioning fix the first failure mode?
Positioning is the deliberate choice of the context in which the offer is understood.
The hero line carries the position. The "this is for" sentence carries the position. The competitor-contrast section carries the position. Without a position, every section reads as a category description.
April Dunford’s five-component framework is the contemporary canon for the work. Competitive alternatives — what the buyer would otherwise do or use. Unique attributes — what the offer has that the alternatives do not.
Value with proof — what the unique attributes do for the buyer, with evidence. Best-fit buyer — the named buyer who gets the most from the offer. Market category — the frame the buyer should place the offer in.
A page with the five components in place reads as an argument the buyer can place. A page without them reads as a category description that could be about anyone. The fix is not a wordsmith pass on the hero. The fix is the upstream decision the hero is meant to express.
For more on the upstream work, see how to write a simple value proposition for a small business.
How does voice fix the second failure mode?
Voice is the register the brand uses every time it talks.
A page without an articulated voice defaults to the median register of the training data. That register reads as no voice at all. The buyer scans the page and forms no impression of the brand because the page gave them nothing distinctive to form an impression of.
A short voice spec is enough to fix the failure mode. Five sentences on how the brand sounds. Three sentences on how it does not sound.
Three banned phrases. The spec fits on a single page. The writer and the tool both have a target to write against.
Five vectors give the spec scoreable axes. Formality. Warmth. Directness.
Humor. Technicality. The brand commits to a position on each axis. The page that ships against the spec carries the writer’s fingerprints because the spec named what those fingerprints are.
The voice spec is the artifact most teams skip. They will write a positioning doc. They will not write a voice spec.
The page that ships without one reads professional and forgettable. The page that ships with one reads as a page from a brand that has decided who it is.
For more on the editing pass that protects voice through the AI-drafting workflow, see how to edit AI-generated content.
How does customer language fix the third failure mode?
Customer language is the actual vocabulary buyers use when they describe the problem the offer solves.
The vocabulary lives in the support inbox. The third-party reviews. The recorded sales calls.
The cold-email replies that said no. The vocabulary almost never lives in the marketing team’s planning docs.
A page that uses the buyer’s vocabulary reads as a page that knows the buyer. A page that uses the marketing team’s vocabulary reads as a page written from the inside out. The buyer recognizes the difference inside two paragraphs.
The fix is small and slow. Read 30 reviews. Read 30 support tickets. Read 30 cold-email replies.
Capture the recurring phrases in a shared doc. The doc becomes the substrate the next AI prompt feeds on. The next draft carries the buyer’s vocabulary because the prompt did.
The same edit that swaps generic adjectives for buyer outcomes also produces the kind of specific phrasing answer engines prefer to cite. Voice integrity and citation quality reach for the same edit. The Toolient case study tracked the convergence on a full product catalog — one rewrite served two metrics.
Other questions worth answering
When does spare wording belong instead of buyer-vocabulary swaps?
Three categories sit outside the standard advice. Luxury fashion, design-forward consumer brands, and high-design service businesses convert on deliberate spareness. The buyer recognizes the identity before the offer is explained. Most other service businesses do not sit inside that carve-out, where spare wording reads as a missing argument.
How often should the buyer-vocabulary research get refreshed?
Every six months for stable categories. Quarterly for categories where the buyer’s situation shifts fast. The recurring sources stay the same: support inbox, third-party reviews, recorded sales calls, lost-deal email threads. A roughly 30-minute pass through fresh entries each cycle catches new phrasing before it spreads through the rest of the brand wording.
What budget should a small-business owner expect for the rewrite work?
Two to four hours of owner time for the diagnostic and rewrite together, plus the writer’s fee if the owner does not draft. The buyer-research substrate is the multi-day investment that sits behind it. Once the doc exists, every future rewrite pulls from it. ChatGPT and Claude both produce a usable draft in 2026 when the prompt carries the position, the spec, and the buyer-vocabulary snippets.
How does translation change the spec work for multilingual brands?
Two layers, not one. Translation forces a spec rewrite alongside the word-level work because the buyer’s vocabulary, register, and identity cues shift across cultures. The cleanest workflow hires a translator who also writes in the target market. Brief them on the original spec and ask them to rewrite it for the new locale before drafting the localized homepage.
How would you diagnose and fix your next generic page?
Three passes, in order.
Pass one. Diagnose the failure mode. Read the page aloud and ask which question the page is failing to answer. Is the problem positioning — the buyer cannot place the offer?
Is it voice — no distinctive register? Is it customer language — the page is full of marketing-team vocabulary? Most pages have at least two of the three. Fix the most-broken one first.
Pass two. Add the named hero line, the named voice spec, or the named buyer-vocabulary swaps the page is missing. Each fix is concrete.
The hero gets the five-component position. The page gets the voice-spec rewrite. The body copy gets the buyer-vocabulary swaps from the support-inbox notes.
Pass three. Read the rewrite aloud against the diagnosis from pass one. Confirm the symptom is gone.
The page that started as a category description now reads as a position. The page that started in the median register now reads in the brand’s voice. The page that ran on marketing-team vocabulary now runs on buyer vocabulary.
Forty minutes per page after the customer-research is done. The page that ships reads as a page that could only be about this business and earns AI citation lift in the same edit.
If your homepage reads professional but forgettable and you cannot tell which of the three failure modes is dominant, you can contact me here. Send the page and three lines from a recent customer email. I will run the diagnosis pass and name the dominant failure mode.
No charge and no follow-up call. The page after the diagnosis is the one that earns the buyer’s second read.