Your AI Sounds Like Everyone Else's. Here Is Why

AI content sounds generic because most of it is built from the same two ingredients, a capable model and public information, and every competitor in your category has access to both. Businesses notice the result and treat it as a tone-of-voice problem. It mostly is not. The sameness buyers actually notice sits in the substance, in the problems a proposal names and the argument an email makes, and that substance can only come from what your business has decided, including what it has decided against.

This post is about that second layer. Why a model with nothing specific to go on falls back on the category average is covered elsewhere on this blog, in what AI agents need before they act and in the direction problem that caps every AI rollout. The question here is what happens when a whole category does that at once, where the cost lands for a commercial team, and what the fix actually requires.

Why does AI-generated content sound generic?

The first answer most businesses reach is voice. The drafts sound like a press release, so someone assembles sample posts and style notes, the model learns the vocabulary, and the output starts to sound like the business. That work is worth doing. Our own product, Semoria, exists because tone of voice turned out to be a solvable engineering problem.

Then the complaint comes back in a different form. The proposal reads well and still does not win. The outbound email sounds like the sender and still gets no reply. Nobody can point at a sentence that is wrong, and yet the work could have come from any firm on the shortlist.

That is because tone was never the whole problem. A draft can match your voice perfectly and still make your competitor's argument, because the argument came from the model's sense of what a firm like yours typically says, not from anything your business decided.

Do competitors get the same output from the same AI?

Increasingly, yes, and the reason is how many of them are now doing it. According to the Office for National Statistics, "the AI technology that businesses had adopted the most was text generation using large language models (17%)" in June 2026, up 12 percentage points since the question was first asked in September 2023. That is across the industries the ONS business survey covers, which leaves out finance and insurance among others. Across the same survey, 29% of businesses now use at least one kind of AI, and the kind they have adopted most is the kind that writes.

The more useful evidence is on what happens to the work. In a controlled experiment published in Science Advances, Anil Doshi and Oliver Hauser had 293 writers produce short stories, some with AI-generated ideas to draw on and some without, and had 600 separate readers judge the results. The AI-assisted stories were rated as more creative and better written, and they were also "more similar to each other than stories by humans alone". The authors describe it as a social dilemma, in which "writers are individually better off, but collectively a narrower scope of novel content is produced".

Short fiction is not a sales proposal, and the study does not claim to measure one. The mechanism transfers, though, and it is uncomfortable for anyone selling in a crowded market. Each firm that adopts AI drafting gets better individual output. The category as a whole gets more alike, and a buyer comparing three proposals sees three competent documents making one argument.

Switching supplier does not obviously help either. A preprint released on arXiv this summer, not yet peer reviewed, asked 18 model configurations from 11 vendors to pick out the sentences that mattered in real web documents, and compared their choices with those of ordinary readers. On the median document each party picked 14 sentences out of 70, and "two readers share 4.1 and two models 8.7". Two models agreed with each other roughly twice as often as two people did, and the authors report the effect held across vendors rather than belonging to any one of them. If that holds up, changing vendor gets you different phrasing and much the same view of what matters.

Where sameness costs a commercial team

It rarely shows up as a complaint about writing. It shows up as a commercial number that drifts the wrong way without an obvious cause.

Where it shows	What the AI builds from public inputs	What only your business knows
Proposals	The standard approach for this kind of project, the usual risks, a methodology section any bidder could submit	The approach you would refuse for this client, and the project that taught you why
Qualification	Company size, sector and budget signals, weighted the way every scoring template weights them	The profile of client you have learned to walk away from, even when the numbers look right
Outbound	A pain point every competitor names for this job title, with a line of personalisation from their LinkedIn page	The problem your best clients actually had when they first called, in the words they used
Pricing and scope	Market-rate packaging and the usual caveats	Why your pricing is shaped the way it is, and which discounts you stopped giving
Claims	What firms in the category typically promise	What you will not promise, and the evidence behind what you will

The middle column is not bad work. That is what makes it dangerous. It passes review because nobody can fault it, and it quietly lowers the win rate because it gives the buyer no reason to choose you over the next firm, whose AI wrote the same thing. AI proposal workflows genuinely save most of the hours a proposal used to take, and the hours are real. What decides whether the faster proposal wins is what the workflow had to read.

What is the difference between tone and thinking?

Tone is how something is said, meaning the vocabulary, the sentence length and the level of formality. Thinking is what gets said, meaning which problem you lead with, what you recommend, what you refuse to promise and who you believe the work is really for. Tone can be learned from a few dozen good examples. Thinking cannot be learned from examples at all, because the examples show the output of a decision and not the decision itself.

Our work with Excellerate Services shows the difference clearly, because polish was never what they were missing. Their team was already producing strong LinkedIn posts every week across the UK, the Middle East and South Africa, in two sectors. Their EMEA Sales and Marketing Director was candid about the gap. "These possibly lacked the thought required to enhance our local and global messaging." The engine we built, around a dozen agents each tuned to a region and sector, took production from roughly 12 hours a week to about 2 hours of oversight. The part that matters here is the change the client actually described. Content is now planned locally and tied to global lead generation and marketing, rather than reacting to whatever the week threw up, so the output carries the business's thinking as well as its voice.

What you rejected is the part nobody else has

If thinking is the missing input, the obvious move is to write down what the business believes. Most firms have a version of this already, in a positioning deck or a set of values, and it rarely helps, because it records the conclusions every firm in the category would also claim.

The more distinctive record is of what the business said no to. Every competitor can say they focus on quality and long-term partnerships. Only you know that you stopped taking projects under a certain size because they never ran to plan, that you lost three deals last year to the same objection and changed your scoping because of it, or that you will not promise a delivery date before a discovery call, and why.

Rejections are the most specific knowledge a business holds, and they are the easiest to lose, because nobody records what did not happen. A won deal leaves a contract, an invoice and a case study. A lost one leaves a line in the CRM saying "price" when the real reason was that the buyer never believed you understood their problem. A client you turned away leaves nothing at all.

That is precisely what a model cannot infer from public text, because it was never public. It is also what makes a proposal read like it came from someone who has done the work before. "We would not recommend a phased rollout here, and this is why" is a sentence no competitor's AI will write for you.

How do you stop AI producing average work?

You give it the decisions that are not average, and you put them somewhere it will read every time. In practice that comes down to three habits.

The first is to write the no down with its reason, at the moment it happens. When a deal is lost, record what the buyer actually said. When you turn a client away, record the pattern you recognised. When a claim gets cut from a proposal, record why it was cut. A bare "rejected" teaches nothing, while "rejected, because the last two projects of this shape overran by a month" is a rule.

The second is to hold those decisions in one current place that every AI tool draws on, rather than in each tool's own settings or in a prompt someone wrote in March. That shared, maintained record is what we mean by an intelligence system, and the rejections are some of the most valuable things in it.

The third is to review AI output against the decisions rather than against the voice. The useful question in review is whether the draft makes an argument you would make, not whether it sounds like you. When it does not, the correction goes back into the record so the next draft starts from it.

The model is the same one your competitors are using. What you control is what it works from, and in a category where everyone has the same model, that is the only variable left that belongs to you.

Practical takeaways

1. Stop treating sameness as a tone problem. If the drafts already sound like you and still do not win, the missing input is your thinking rather than your voice.
2. Assume your competitors have the same model. Anything the AI can produce from public information is available to every firm on the shortlist.
3. Record the no, with the reason. Lost deals, turned-away clients, dropped approaches and claims you will not make are the most distinctive knowledge you hold.
4. Put decisions where every tool reads them. One current record beats a good prompt in each tool, because the prompt goes stale and nobody updates it.
5. Review for the argument, not the adjectives. Ask whether the draft makes a case you would make, and write every correction back.

Frequently asked questions

Why does AI-generated content sound generic?

Because a model with nothing specific to work from falls back on what is typical, and what is typical for your category is exactly what your competitors are also getting. The usual fix is to work on tone of voice, which makes the output sound more like you without making it think more like you. The generic quality most buyers notice sits in the argument, in which problems get named, which options get recommended and which claims get made, and that comes from the decisions your business has made, which no public model has ever seen.

Do competitors get the same output from the same AI?

Close to it, if they give it the same kind of inputs, and the inputs are usually the same kind. Two firms in one sector prompting a model about the same buyer, from the same public information, will get drafts that differ in wording and agree in substance. Research on AI-assisted writing has found that work produced with AI help is more similar across writers than work produced without it, and early research comparing models from different vendors suggests that switching supplier does not escape the effect. The difference has to come from what you feed it.

What is the difference between tone and thinking?

Tone is how something is said, meaning the vocabulary, sentence length and formality that make a draft sound like your business. Thinking is what gets said, meaning which problem you lead with, what you recommend, what you refuse to promise and who you think the work is for. Tone can be learned from a few dozen good examples, and tools do it well. Thinking has to be written down, because it lives in decisions your business made and mostly never recorded, and without it a perfectly on-voice draft still makes your competitor's argument.

How do you stop AI producing average work?

Give it the decisions that are not average, starting with the ones that record a no. Write down the clients you turn away and why, the deals you lost and the real reason, the claims you will not make, and the approaches you tried and dropped. Those rejections are the most specific knowledge a business holds and the easiest to lose, because nobody records what did not happen. Keep them somewhere current that every AI tool reads before it drafts, and review what it produces against them.

Why does an AI-drafted proposal read like every other bidder's?

Because it is usually built from the same material as the other bidders' proposals, namely the tender, the prospect's website and a model's sense of what a good proposal in your sector looks like. That produces a competent document making the category's standard argument. What wins is the part only you could write, such as the approach you would not take for this client and why, the risk you have seen sink a similar project, and the reason your pricing is shaped the way it is. If that reasoning is not somewhere the proposal workflow can read, it cannot appear in the draft.