Table of Contents
TL;DR
- A persona document is research. An ideal-client profile is a decision. Most small businesses with bad copy have a positioning problem the persona was supposed to fix and did not.
- An ideal-client profile fits in three sentences. Who the buyer is by job and situation. What the buyer is hiring the product to do. Who the buyer is not.
- April Dunford’s deliberate-exclusion discipline names who the product is not for. The exclusions free the rest of the positioning to be specific.
- Joanna Wiebe’s three-filter test catches the most common failure — the ICP that describes a category instead of choosing one. Specific, outcome-named, believable.
- Answer engines summarise long pages into short recommendations. A sharp ICP survives the summary — a vague ICP gets averaged into the category description and disappears.
Most small businesses with bad copy do not have a copy problem. They have a positioning problem the persona document was supposed to fix and never did.
The 20-page persona sits in a Word doc that no one opens after the workshop. It describes the buyer’s age, marital status, podcast preferences, and weekend hobbies. It does not change a single sentence on the homepage.
The ideal-client profile is the opposite shape. Three sentences. Fits on an index card. Reads like a choice rather than a description.
That distinction is the whole post.
Why does the 20-page persona document not help your copy?
A persona is a research artifact. An ideal-client profile is a copy decision.
The two get conflated because both involve writing about buyers. The persona reads like ethnography — demographics, psychographics, day-in-the-life. The ICP reads like a wager — this buyer, not that one, for this reason.
Most agencies that run persona workshops produce real, useful research. The output is also rarely opened again after the printout reaches the client. The homepage that came out of the workshop reads like the homepage that went in.
The pattern is consistent enough across small businesses to name. Your copy converts on the choices the page makes, not on the research that preceded it. The persona describes a population.
The ICP names a person. Pages convert when they are written for one person.
The fix is not a longer persona. The fix is the harder question. Which buyer am I writing for, and which buyer am I willing to lose?
What does an ideal client actually look like, in three sentences?
Three sentences, on a card.
Sentence one names the buyer by job and situation. Not "small business owners aged 25-55." Closer to "the solo accountant with eight tax-season clients and no marketing budget" or "the e-commerce founder running one Shopify store from a kitchen table."
Sentence two names what the buyer is hiring the product to do. Not the feature list. The job. "Get the year-end books into a state where the tax filing takes one Saturday, not three." "Make the homepage stop reading like every competitor’s homepage so referrals convert into customers."
Sentence three names who the buyer is not. The exclusion. "Not for venture-backed teams who already have a marketing director." "Not for hobby bloggers who do not need this to pay rent."
Three sentences. Reads like a choice. Could not appear on a competitor’s site without changing the meaning of every other sentence on the page.
Why do exclusions sharpen the ICP more than another round of demographics?
April Dunford built the deliberate-exclusion discipline into her five-component positioning framework, alongside competitive alternatives, unique attributes, value and proof, and market category. The exclusion is the fifth lever and the most often skipped.
Naming who the product is not for is uncomfortable. Founders worry the page will lose buyers it might otherwise win. The opposite is closer to the truth. A page that tries to win every buyer wins none of them clearly enough to compete.
The exclusion exercise forces a harder question than the inclusion one. Who is this product worse for than the obvious alternative? The answer almost always names a real category — enterprise teams who need a procurement contract, hobbyists who need a free tier, agencies who need white-label resale. Once that category is named on the page, the rest of the page is free to be specific for the buyer who remains.
Naming who you serve is half the work. The other half is deciding who you exclude and saying it out loud. The page that turns the wrong buyer away earns trust from the right one.
How does the Jobs-to-be-Done frame pull the ICP up one altitude?
Clayton Christensen and Bob Moesta argued that buyers do not buy products because of attributes. They hire products to do specific jobs. The reframe sounds small. The implication is large.
The classic example is the McDonald’s milkshake. The brand assumed buyers wanted thicker chocolate or fresher fruit. The interviews surfaced something else.
Half the morning sales went to commuters with a long solo drive ahead. The milkshake was thick enough to last forty minutes, easy to drink one-handed, filling enough to skip a snack at ten.
The buyer was not "men aged 25-45 who eat fast food." The buyer was "anyone with a forty-minute solo drive and no time for breakfast."
That is the altitude shift. Demographic ICPs name a category the buyer happens to belong to. Job-aware ICPs name the moment the buyer reaches for the product.
Moments produce specific copy. Categories produce generic copy.
Most small-business homepages live one altitude too low. They describe the buyer’s job title or industry, and the page reads like the trade-association brochure. The ICP that names the moment writes the page that converts.
What does the three-filter test catch in most ICP statements?
Joanna Wiebe proposed three filters for any value statement, and the same three filters work on the ICP itself. Specific. Outcome-named. Believable.
Specific means the sentence could only be said about this product, this buyer, this situation. Most ICPs fail this filter. They describe a category that fits dozens of competitors.
"Small business owners who want to grow" is not an ICP. It is a market.
Outcome-named means the buyer recognises the result they wanted in the wording. Not "we serve restaurateurs" but "we cut your weekly inventory count from four hours to forty minutes." The buyer reads the second version and recognises the moment.
Believable means the buyer does not flinch. Promising "ten times your revenue" to a solo consultant fails the test. So does promising "enterprise-grade workflow" to someone whose enterprise is a kitchen table.
Run the three filters on the next ICP draft. If the same words could land on a competitor’s page without changing meaning, the ICP has been described, not chosen.
How does AI search reward a sharp ICP and erase a vague one?
ChatGPT, Claude, Perplexity, and Gemini compress long pages into two-paragraph recommendations and three-bullet lists. The summary keeps what is specific and discards what is generic. That mechanism rewards a sharp ICP and erases a vague one.
A page written for "the solo accountant with eight tax-season clients" carries a specific job, a specific situation, and a specific exclusion. The chat surface returns the page with the buyer description intact, because there is something to return.
A page written for "small business owners who care about growth" gets averaged into the category description. The chat surface returns "small business advisor" with no specifics, because the page contributed no specifics.
A cmswire 2026 piece on AI customer experience put the boundary in plainer language. "The first letter in AI stands for artificial," the piece argues, and what customers reach for under stress is authenticity. Generic ICP language reads as artificial because it was averaged across competitors. Specific ICP language reads as authentic because it was chosen.
The reader recognises the difference inside two sentences. So does the engine. They reward the same sentence for almost the same reason.
For more on positioning above the AI rather than alongside it, see position above AI versus position alongside AI.
What does an ICP shift look like in a small business?
Founders often build for buyer A and discover that buyer B is the actual best-fit customer. The marketing keeps speaking to buyer A. Growth stalls.
The diagnostic lives in three places, read side by side.
The support inbox tells you which questions arrive most often, from which kind of buyer. If the bulk of the questions come from a segment the homepage does not name, the ICP has shifted. Read the last sixty support tickets and group them by sender type, not by topic.
The customer-interview transcripts tell you who describes the result in the strongest words. Three short interviews, twenty minutes each, with the customers most clearly happy. The buyer who uses the most concrete language about the change the product produced is the buyer the next homepage should be written for.
The accounting export tells you which segment carries the highest lifetime value and the lowest cancellation rate. Sort the customer table by total revenue, then by months retained. The top-decile rows describe a real ICP that may not match the homepage’s stated audience.
When all three sources point to the same buyer, and that buyer is not the buyer the page was written for, the ICP has shifted. The rewrite is overdue. For more on identifying the pain points the new ICP names, see how to find the customer pain points AI summaries miss.
Other questions worth answering
How do you draft the line that names which buyers the product turns away when no customer interviews exist?
Three sources, drawn from April Dunford’s exclusion discipline. List three buyers you would politely turn away tomorrow. Then list three use cases the product handles worse than the obvious alternative.
Then list three nearby categories the product does not belong in. Read all nine together. In cmswire’s February 2026 analysis, ‘specific over averaged’ is the consequence — the strongest of the nine writes the line.
Should the buyer profile sit on the homepage, or stay internal as a writer’s tool?
No — the ideal-client profile is a writer’s tool, not a customer-facing artifact. In April Dunford’s framework, the canvas sits in the writer’s notebook, never on the homepage.
The buyer never sees ‘our ideal client is the solo accountant with eight tax-season clients’ written anywhere on the homepage. They read a paragraph that only makes sense to that accountant.
What is the difference between a positioning problem, a messaging problem, and a tactics problem?
Three problems, three fixes. A positioning problem means the buyer cannot place the product or its audience — the question April Dunford built her framework to answer. A messaging problem means the value prop is correct but the words do not land. A tactics problem means the headline or CTA is weak.
Tactics fixes take an afternoon. Messaging fixes take days of customer interviews. Positioning fixes take weeks.
How often should the buyer profile be reviewed after the first draft lands?
Two reviews a year on the calendar, plus three triggers. By 2026 the practical approach pairs Dunford’s positioning canvas with three running diagnostics: the support inbox, customer-interview transcripts, and the accounting export. Trigger one is a noticeable shift in inbox themes.
Trigger two is a quarterly accounting review where the highest-lifetime-value customers do not match the homepage’s stated audience. Trigger three is a launch into a new product line.
Which two questions should you answer before your next page rewrite?
Two questions, in order, on a single index card.
Who is this product not for, named by job and situation? Write the exclusion in one sentence. Resist the temptation to soften it.
What is the buyer hiring this product to do, named in the buyer’s own words? Pull the language from a recent customer interview if you have one, from the support inbox if you do not.
Read the next homepage paragraph against the card. If the paragraph could appear on a competitor’s site without changing meaning, the ICP has not yet earned the page. If the paragraph names the job, names the exclusion, and survives a Joanna Wiebe filter test, the rewrite has somewhere to land.
The card is the work. The page is the artifact.
If you have a homepage that is not converting and you suspect the ICP behind it was averaged into a category description, you can contact me here. I will read the page against a sharper ICP that a thirty-minute conversation can usually surface. There is no charge, no pitch, and no upsell.
The card is small enough to write in a morning. The page is the part that takes the rest of the week.