Why the Result Reads Like Everybody Else's

What likely means

The tool is producing, word by word, the continuation that is most probable given everything before it.

Which means it converges on the most usual way of saying anything, because the most usual phrasing is the most probable one.

That is not a defect being worked on, it is the mechanism, and it explains everything about how the output reads.

The average of everything written about your trade is exactly what it produces, and the average is what your competitors' pages already say.

The consequences follow directly

• Common phrasing over precise phrasing.
• Middle positions over strong ones.
• Everything covered to equal depth.
• No preferences or opinions.
• Cliches, because cliches are probable.

The third is worth noticing, because it is exactly the uniform coverage that marks a page written to occupy a position rather than to answer somebody. A person who does the work writes unevenly, spending three paragraphs on what matters and a sentence on what does not.

Distinctive means improbable

The point that makes this structural rather than solvable.

Everything that makes a page worth reading is unusual: your specific figures, your particular exception, the job that went differently, the opinion most of your trade disagrees with.

Unusual is the opposite of probable, which is what the method selects for.

So the tool cannot generate the things that differentiate you, not because it is not good enough yet, but because differentiating means being unlike the average and it is producing the average.

That is a useful thing to hold onto, because it stays true regardless of how much these tools improve at everything else.

Why this matters more this year

Since August changed the value of being ordinary.

The content update was explicitly about pages that could have come from anywhere, made to fill a position rather than to help somebody.

Generated text is that description in its purest form, produced by a method that optimises for exactly what was named.

So the timing is unfortunate for anybody planning to publish volume this way: the tool arrived four months after the industry stated plainly that it would be discounting the output.

Which does not mean generated text is detectable or penalised as such, and it does mean that content indistinguishable from everybody else's is now explicitly worth less than it was.

A worked example

A business generated five pages and read them alongside five they had written themselves two years earlier.

The generated ones were better written: cleaner sentences, better structure, no awkward phrasing.

The written ones contained a price, a photograph of a failure, an argument about why a common approach is wrong, and a paragraph about a job that went badly.

Every one of those four had produced customer conversations, and none of them could have been generated because none was in the average.

Their conclusion was that the tool wrote better sentences and had nothing to say, which is a fair summary.

It has no way to prefer one thing

A related limitation that explains the blandness people notice and cannot name.

Asked which of two approaches is better, it will describe both fairly and decline to choose, because the average of everything written on the question contains both positions.

Whereas a customer reading your page wants somebody who has done this many times to say which one they would use and why.

A recommendation is the single most valuable thing an experienced person can offer, and it is precisely what an average cannot contain.

Which means the page it writes will always end with it depends on your circumstances, and yours should not.

What to do with your own writing

Since the same argument applies to how you write regardless of any tool.

Ask of each paragraph whether a competitor could publish it, and cut anything where the answer is yes and it adds nothing.

Add the thing you would not expect to see elsewhere: a figure, an exception, a disagreement, a case where you were wrong.

Take a position rather than describing the options evenly, since a page that recommends nothing is the shape the average produces.

Write unevenly on purpose, spending the space where the real difficulty is.

That advice was true before this week and is simply easier to explain now.

The awkward implication

Worth facing rather than avoiding.

If a machine with no knowledge of your business can produce something close to your current pages, those pages were not saying much.

That is uncomfortable and it is useful information, obtainable in ten minutes.

Generate a version of your own most important page and compare it honestly with what is live.

Where the difference is small, the problem was there before this tool existed and is now visible.

The sameness compounds

A second-order effect worth seeing early.

If a large share of what gets published from now on is produced this way, the average that these tools are trained on becomes more uniform over time.

Which means the ordinary version of any subject gets more ordinary, and the gap between generated pages and pages written by somebody who does the work widens rather than narrows.

That is speculative and it points in a useful direction: the value of a specific, first-hand, slightly awkward account of real work is more likely to rise than fall.

Which is an argument for continuing to write the way you already should have been, rather than for anything new.

The counter-case

Ordinary writing has its place.

A great deal of what a site needs is genuinely standard: what you do, where you are, how to get in touch, what happens next.

For that, average phrasing is correct and being distinctive would be a distraction.

These tools are also improving quickly, and confident claims about what they will never do have a poor history.

Use the average where average is right, supply the unusual yourself, and check your own pages against a generated version.

What to do

1. Generate a version of your best page.
2. Compare it honestly with what is live.
3. List what only you could have written.
4. Add more of that.
5. Cut what a competitor could publish.
6. Take a position rather than describing options.
7. Write unevenly where it matters.

Step two is the uncomfortable one and the informative one, since a small difference means the page was already saying very little.

The failure this describes is covered in pages written for search engines.

Frequently asked questions

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.