How algorithms rewrite the way we speak
Hey reluctant copy machines—
Have you noticed how often you open a blank box and you’re not really thinking about people anymore?
Not “What do I actually want to say?” But: “How will this perform on LinkedIn?” “Will Google like this title?” “Will the AI summarise this nicely?”
You’re about to write to your friends, your clients, your readers. And you start negotiating with a sorting system instead.
At some point we stopped talking to humans and started talking to algorithms that stand between us and them. And like any power that stays long enough in the room, those algorithms have started to change our voice.
This is an article about that slow mutation.
The small betrayal before every “post”
There’s a specific micro-moment that keeps haunting me.
You’re staring at the “Share your thoughts…” box.
You type a sentence that sounds like you. It’s slightly weird, maybe too long, maybe a bit dark. The kind of thing you’d actually say over coffee.
Then you pause.
You picture the feed. The reach. The algorithmic god you need to please.
And in your head you hear the internal product manager:
“Can we make this more actionable? Maybe: 5 lessons I learned from…?”
Delete. Rewrite. Sanitise.
You add a safe hook. You sprinkle “value”. You drop a question at the end to “spark engagement”.
It still technically says what you meant. But now it sounds like a cousin of all the other things on that platform. Not exactly fake—just… optimized to death. X
We’ve learned to do this so often that we barely notice the betrayal anymore.
A short history of learning to talk to machines
This didn’t start with generative AI. We’ve been negotiating our language with software for decades.
First came early SEO: pages stuffed with keywords, titles that read like spammy incantations, articles that existed less for humans and more for Google’s crawler.
Then social feeds: we discovered what “works” on each platform and slowly grew dialects around it.
- Twitter taught us the sharp one-liner, the thread, the hot take.
- Instagram trained us in caption minimalism plus a side of hashtags and soft vulnerability.
- LinkedIn engineered that particular tone: humble brag, life lesson, invitation to “drop your thoughts in the comments”.
Now we’re in the AI-assisted phase. We don’t just think of Google or the feed. We think:
- “Will this be easy to summarise by an LLM?”
- “Will this appear nicely inside an AI Overview?”
- “Can I ask a model to rephrase this ‘in a more LinkedIn style’ before posting?”
The machine is no longer just at the door, deciding who comes in. It’s in the room, suggesting how you decorate your sentences.
Algorithmic dialects: the accents of platforms
Spend ten minutes scrolling anywhere and you’ll hear them.
Each platform has developed an algorithmic dialect: a way of sounding that the system quietly rewards and users end up imitating.
LinkedIn has its particular music: “I used to struggle with X… then something happened… here are 3 lessons… what about you?” Even the “brutally honest” posts often use the same skeleton.
YouTube titles shout in a specific voice: “I tried X so you don’t have to”, “Nobody tells you this about…”, “I did Y for 30 days—here’s what happened”.
TikTok compresses everything into hyper-dense hooks and private jokes. If you don’t speak the dialect, you feel old in five seconds.
Email newsletters and SEO blogs have their own template: keyword-friendly headers, numbered lists, respectful of what the crawler likes to see.
And floating above all of that, there’s now AI-speak: that neutral, over-explanatory, slightly beige tone you get when you click “rewrite” too many times. It’s not wrong. It’s just not anyone.
None of these dialects emerged organically from a community of writers. They emerged from metrics.
The algorithm will not tell you “Talk like this”. It just shows you which shapes get more reach, watch time, clicks. We learn. We adapt. We copy.
And slowly, the language that makes it through the filter starts to reshape the language we use everywhere else.
Here’s the loop, more or less:
- A platform decides what matters: clicks, dwell time, watch time, comments.
- The recommendation system learns that certain words, structures, rhythms keep people there longer.
- Creators adjust their style to match what survives.
- Audiences get used to those shapes and start imitating them in their own communication.
- AI models get trained on that entire mess and spit it back as “good writing”.
So you end up in this slightly absurd situation:
Humans imitate algorithm-friendly language. —> AI imitates humans imitating algorithms. —-> Then humans copy the AI output because “it sounds right”.
When you stare at a corporate blog, a LinkedIn feed or a stack of AI-generated landing pages, the feeling you get is often the same: a huge, smooth surface of text that could have been written by any mid-level marketer on Earth—or by any model.
And the things that don’t fit the pattern?
- Ambiguity,
- long arguments,
- jokes that take time to land,
- anger that doesn’t resolve into a lesson,
- grief that doesn’t convert into a call to action—
those often fall through the cracks, because they don’t “perform” in the right way, in the right time window, on the right graph.
What doesn’t light up the dashboard is slowly exiled from how we speak in public.
What we lose when we all sound “platform-ready”
There are a few different losses happening at once.
Cognitive
Thinking starts to take the shape of the formats we use most.
We break ideas into blocks because blocks can be:
- scrolled,
- skimmed,
- A/B tested.
We compress complexity so it fits inside a screen-capture carousel or a 30-second video.
We start making decisions not based on “what is true?”, but on “what can be said quickly and still make sense on this platform?”.
Political / social
Arguments that are simple, polarising, easily meme-able win the oxygen.
Things that require time, context, and a slow burn fall behind. There isn’t much room for “I don’t know yet” or “it’s complicated” in an environment built around instant reaction.
The algorithm isn’t evil. It’s just blind to anything that doesn’t translate into measurable engagement. But that blindness has consequences.
Aesthetic
Voices converge.
Brands that swear they want to be “unique” all end up speaking in some variation of the same growth-hacked, AI-polished safe tone.
Writers who had sharp edges file them down because edges reduce reach. The result is a giant soup of correct, smooth, unmemorable language.
Personal
This is the part that bugs me the most.
You start reading your own emails, your own about page, your own posts and you hear… the platform.
Not you. Not even your industry. The platform.
Your sentences are fine. Your arguments are valid. But if I swapped your name with someone else’s, nothing in the voice would protest.
AI training on our most optimised selves
Now put LLMs into this picture.
We ask models to:
- “rewrite this to sound more like a LinkedIn post,”
- “make this more SEO-optimised,”
- “turn this into a high-converting landing page.”
Those outputs are then pushed into the wild. Some perform well. The good performers get more visibility, more links, more reuse.
In the next training round, models see more of that. They learn that this is what “professional”, “engaging” or “authoritative” looks like.
So the system ends up feeding on its own echo:
- Humans optimise for algorithms.
- AI trains on that optimised language.
- Humans copy AI because it “sounds right”.
At each loop, the range of accepted voices narrows a little.
It’s not a conspiracy. It’s just what happens when feedback loops are left alone long enough.
How to write for machines without becoming one
I don’t think the solution is to pretend algorithms don’t exist.
Ignoring SEO doesn’t make you noble, it just makes you invisible. Refusing to think about how feeds work doesn’t protect your voice, it just makes sure nobody hears it.
The question is different:
How do we make our language legible to machines without letting machines dictate all of it?
One way to think about it is two-layer writing.
Layer one: structural clarity
You keep the parts that help both humans and machines understand what’s going on:
- a clear sense of topic,
- headings that actually say something,
- paragraphs that follow a logic rather than wandering aimlessly.
You avoid deliberate obscurity—not because of the algorithm, but because confusion rarely equals depth.
This is the layer that makes your work:
- findable,
- quotable,
- indexable.
Layer two: human texture
Inside that structure, you reclaim all the things that don’t particularly help a ranking system, but absolutely help a human:
- metaphors that belong to you,
- jokes that risk not landing,
- sentences that change rhythm mid-way because that’s how you think,
- stories that don’t resolve into “3 key takeaways”.
You accept that some of this will be invisible to the metrics and still worth writing.
On the surface, the article might look slightly “SEO-aware”. Underneath, the voice is not negotiable.
Reclaiming a few rooms where you don’t talk like this
There’s one more move that has nothing to do with strategy and everything to do with sanity.
Leave parts of your life un-optimised.
Write some things that will never be:
- ranked,
- summarised,
- turned into a carousel,
- parsed by a model.
Give yourself small private rooms where you don’t talk like a brand, a creator or a growth marketer:
- the email to a friend that isn’t “building your personal brand”,
- notes to yourself that sound exactly as messy as your head,
- a document where you allow paragraphs that are too long for LinkedIn and too slow for Twitter.
If every sentence you produce sounds platform-ready, then maybe none of it is truly yours anymore.
Algorithms are not going away. AI models are not going away. Neither is the temptation to let them quietly run your vocabulary.
But you still get to choose where the negotiation stops.
You can make your work discoverable without donating your entire voice to the metrics.
Until next time, stay unpolished.
Alex
If your brand sounds more and more like “the internet in general”, it’s not a coincidence. You’re probably optimising for the same dashboards as everyone else.
At Kredo Marketing, we help companies design communication that:
- is legible to algorithms (SEO, feeds, AI),
- but still sounds unmistakably like them.
If you want to stress-test your tone of voice, your website copy or your LinkedIn presence against this “algorithmic dialect” lens, let’s sit down and dissect it together.
