Brand and Positioning 12 min read

Your website should turn people away

The advice is sound. The standard execution inverts all three conditions the research says the effect depends on.

Somebody told you to niche down. So there is now a line near the top of your homepage saying you are not for everyone, or that you are probably not the cheapest, or that if a client wants a quick logo they should look elsewhere.

It felt bold when you wrote it. It has not obviously done anything.

The advice underneath it is sound and older than any of us. A page that tries to lose nobody gives nobody a reason to choose it. What almost nobody tells you is that there is research on where the disqualifying line goes, how large it should be, and what it should be about, and that the standard execution gets all three of those wrong.

In this article

Why admitting a limit works at all

Martin Eisend published a meta-analysis of two-sided advertising in the International Journal of Research in Marketing in 2006, pulling together the empirical work on messages that disclose something negative alongside the positive claims.

Two-sided messages significantly raise the perceived credibility of the source and reduce negative cognitive responses in the reader. The largest effect in the analysis was on perceived novelty. Saying something unflattering about yourself makes you both more believable and more interesting, and the second of those is doing more work than most people assume.

The mechanism is not complicated. Every business in your category claims to be thorough, responsive and genuinely interested in outcomes. Those claims cost nothing to make, so they carry no information. A claim only tells a reader something if making it was expensive, and admitting a limitation is expensive because it might lose you the sale.

Which is why your reassurances slide off people. Not because they are untrue. Because the reader has no way to tell them apart from the identical reassurances on the four other tabs they have open.

Eisend found one more thing that matters here. Voluntariness affects the outcome. A disclosure the reader knows you chose to make reads differently from one you were obliged to make, which puts a limit on how far you can borrow this effect by publishing terms and conditions nobody asked for.

The dose and the placement

This is where the standard advice comes apart.

Danit Ein-Gar, Baba Shiv and Zakary Tormala published four studies in the Journal of Consumer Research in 2012 on what they called the blemishing effect. A small dose of negative information added to an otherwise positive description can leave people more favourably disposed towards the thing being described.

The word doing the heavy lifting is small. The effect is bounded by two conditions the authors are explicit about.

The first is processing effort. It emerges when the information is processed effortlessly rather than effortfully. Someone skimming, distracted, or deciding quickly shows the effect. Someone reading carefully and weighing every attribute does not.

The second is order. The negative has to follow the positive rather than precede it. When it comes afterwards, readers evaluate the preceding positives more favourably and discount the minor negative. When it comes first, that does not happen.

Eisend’s meta-analysis supplies the third constraint. The relationship between the amount of negative information and attitude towards the brand is curvilinear. There is an optimum, past which more disclosure works against you. Nothing in this literature says that being harder on yourself is better.

Now look at the standard execution again. A prominent disqualifier, high on the page, about a central attribute like price or commitment, stated forcefully. That is a large dose, placed first, on the thing the reader cares most about. Every one of the three conditions is inverted.

Two of these findings do also cut against each other in a way worth naming. The blemishing effect wants a minor blemish on something peripheral. The credibility logic wants a disclosure that costs you something, which usually means a central attribute. Those pull in opposite directions, and the research does not resolve it for you, because the studies were run on product descriptions rather than on the question of who a service is for.

The reader who is not a person

Everything above assumes a human lands on the page. Increasingly the first thing to read your site is a model, summarising you for somebody who asked a question and never arrived.

Google’s guide to optimising for generative AI features describes the mechanism. Queries fan out: the model generates a set of related sub-questions and retrieves pages to ground its answer against. Their worked example takes “how to fix a lawn that’s full of weeds” and fans it into “best herbicides for lawns”, “remove weeds without chemicals”, and “how to prevent weeds in lawn”.

Apply that to somebody looking for what you sell. They ask one question. The model asks eight, and several of them are sorting questions. Who does this for businesses my size. Who works this way rather than that way. Who is going to be a poor fit.

A page saying you are not for everyone answers none of those, because it contains no fact to retrieve. A page saying you decline projects under six weeks answers one of them directly.

Google names this distinction themselves, in the same guide, and their framing is unusually blunt. They contrast commodity content, which they describe as based on common knowledge that could have originated from anyone, with non-commodity content that carries a specific expert or experienced take beyond the ordinary. Their commodity example is an article titled seven tips for first-time homebuyers. Their non-commodity example is one about waiving a building inspection and what was found in the sewer line.

Every generic disqualifier is commodity by that definition. It could have come from any business in your category, which means it distinguishes you from none of them, to a reader or to a retrieval system. The specific one is the opposite, and it is specific precisely because it cost you something to write.

Here is where the popular response goes wrong, and Google says so directly. Their own list of things to ignore includes overfocusing on structured data, and states that no special schema.org markup is needed for generative AI search and that structured data is worth keeping for rich results eligibility rather than for AI visibility. The same list rules out llms.txt files, breaking content into chunks, and rewriting pages specifically for AI systems.

Their instruction instead is to organise content in a way that helps your readers, write for the human audience, and make pages for that audience rather than for generative AI search.

So there is no separate machine-facing version of your website to build. The sentence that tells a model who you are wrong for is the same sentence that tells a person, sitting in the same paragraph, doing both jobs at once. That is a relief in workload terms and an uncomfortable finding in every other way, because it removes the technical task people reach for instead of the harder one.

It does create a genuine tension with the placement rule though. Retrieval wants the fact stated plainly and unambiguously. The blemishing effect wants it small and late so a scanning human reads past it favourably. Those are different demands on the same sentence.

The resolution is that a page can carry the fact twice in different registers. Once inside the flow of the argument, late and proportionate, where a person meets it. Once as a plain question and answer at the end, where it is unambiguous and easy to lift. A frequently asked questions section is not decoration for this reason. It is the place where the specific statement can be flat and explicit without disrupting how the page reads to somebody scrolling.

One thing worth knowing about the next layer. Google’s guide notes that browser agents acting on someone’s behalf inspect the rendered page, the DOM, and the accessibility tree. Which means alt text, heading order and semantic markup are read by more than screen readers now. Nothing to build there either, but a reason to stop treating accessibility as an optional pass at the end.

What this means for an actual homepage

Start with the condition you already satisfy. Homepage scanning is a low-effort processing environment almost by definition. Nobody arrives at a services page in a careful, attribute-weighing frame of mind. They arrive with four tabs open, half-deciding. That is the exact state the blemishing effect requires, which means your homepage is better suited to this than a proposal or a pitch, where the reader is concentrating and the effect disappears.

Then place it late. The disqualifying line belongs after you have said what you do and who it is for, not above it. This is the change that costs nothing and that almost everyone gets backwards, because a bold opening disclaimer feels braver than a quiet one further down, and bravery is not the variable being measured.

Keep the dose small. One specific limitation lands better than a section of them. A page listing everything you are not reads as positioning theatre, and the curvilinear finding says you pass the useful point and keep going.

Make it specific and checkable. “We are not for everyone” is not a disclosure, because it costs nothing and disqualifies nobody. “I do not take on projects under six weeks” costs you something real and tells a reader something they can act on. The first is a posture. The second is information.

Aim it at fit rather than at quality. A limitation about how you work, what you will not do, or what a client needs to bring gives the reader a genuine sorting criterion without asking them to accept that your service is worse. Price is the hardest one to place, because it is central for most buyers, which is where the tension between the two findings sits.

Underneath all of this is a prior question the research cannot answer for you. Every one of these moves requires you to already know what you refuse to do, and that is not a copywriting decision. If your disqualifier is vague, that is usually because the position behind it is vague, and no amount of rewriting the sentence will sharpen it. This is the same reason redoing the design does not fix a site that does not sound like you.

Which is also why the borrowed version fails so reliably. Copying a competitor’s disqualifying line gives you their boundary rather than yours, and a boundary that is not actually yours cannot be defended when an enquiry arrives that crosses it. You will take the job. The line on the page then becomes something your website claims and your business does not do, which is the borrowed voice problem in its most expensive form.

The test for whether you have a real one is simple enough. Think of the last enquiry you accepted and should have turned down. Whatever made it wrong is your disqualifier, already discovered, already paid for. It does not need inventing. It needs writing down, placing late, and keeping small. If nothing comes to mind, the honest reading is that your boundary has not been set yet, and the Brand Mirror is eight questions that put your answers next to what your site currently says.

The point of turning people away was never to seem confident. It is that a reader who cannot find a reason to rule you out cannot find a reason to choose you either, and the two are the same act performed in the same ten seconds.

But you cannot afford to turn work away

This is the objection that actually stops people, and it deserves a straight answer rather than encouragement.

The fear is that a disqualifying line reduces the number of enquiries. It might. What it more reliably changes is their composition, and those are different quantities with different consequences.

An enquiry from someone who was always going to balk at your price does not convert. It consumes a call, a proposal, and a fortnight of intermittent thinking about whether to discount. An enquiry that arrives having already read what you will not do and pressed send anyway has cleared a hurdle before you have spent anything. The first kind is not revenue. It is unpaid work that ends in a polite email.

You can check this against your own records rather than taking anyone’s word for it. Count the enquiries you responded to last quarter and count the ones that became paid projects. Then look at what the unconverted ones had in common. Most solopreneurs find one recurring mismatch sitting under most of them, and that mismatch is the disqualifier they have not written down.

There is a second cost that never appears in a conversion rate. A wrong-fit client who does hire you occupies the calendar space of a right-fit one, and produces work you would rather not be judged by. Your portfolio is assembled from who you said yes to, which means every wrong yes shapes what the next enquiry expects you to be.

None of that argues for turning away work you need. It argues that the enquiries a specific limitation removes are disproportionately the ones costing you time without producing income, and that the arithmetic is available in your own inbox rather than in anyone’s advice.

The genuine risk sits elsewhere. A limitation you have not tested may be wrong, and you will not discover that from a homepage. It is worth stating one and watching what stops arriving for a quarter before you commit to a page full of them. The literature on dose says the same thing from the other direction, which is that one is close to the useful amount anyway.

Common questions

Should my website say who it is not for?

Yes, with conditions. Eisend’s meta-analysis found that messages disclosing something negative alongside positive claims raise source credibility and perceived novelty. The effect is bounded though. Ein-Gar and colleagues found the benefit appears when the negative is small and follows the positive information, and Eisend found the relationship between amount of disclosure and brand attitude is curvilinear, so more is not better.

Where should a disqualifying statement go on a page?

After you have said what you do and who you serve, not before. The blemishing effect emerged when negative information followed positive information rather than preceding it. Readers evaluated the earlier positives more favourably and discounted the minor negative when the order ran that way, and did not when it was reversed.

Does saying “we’re not for everyone” work?

It does very little, because it costs nothing to say and disqualifies no one. A disclosure carries weight when making it is expensive. Something specific and checkable, like a minimum project length or a type of work you decline, gives a reader a sorting criterion they can act on.

Does turning people away help with AI search?

Indirectly, and not through markup. Google’s guidance describes queries fanning out into related sub-questions that the model retrieves pages to answer, several of which are sorting questions about who a business suits. A specific disqualifier is a retrievable fact that answers one. A generic one is what Google calls commodity content, common knowledge that could have originated from anyone. Their same guidance says no special schema is needed for generative AI search and that pages should be written for the human audience.

Can turning people away hurt my business?

The research suggests it can if the dose is wrong. Eisend found a curvilinear relationship between the amount of negative information and attitude towards the brand, meaning there is a point past which further disclosure works against you. The evidence supports one specific, well-placed limitation rather than a section of them.

Keep reading

More on Brand and Positioning