Lab Notes · Concluded

One draft, two prompts, three AI models. Did they improve my writing?

I gave ChatGPT, Claude, and Gemini the same unfinished article to edit, first with a general request to improve it, then with specific journalism and editing instructions. I compared the six responses to see what improved, what got lost, and whether I would publish any of them unchanged.

I found useful edits under both prompts and more I wanted to keep in the Editorial responses. Every version still contained decisions about meaning, structure, or voice that I would revisit. This Lab Note records the experiment behind Writing with AI? Think like a journalist. Give people a reason to read.

Read the main article →

What I wanted to find out

Would a detailed editorial brief produce a better edit than a general request to improve a text? And would either result be something I wanted to publish?

I wrote the beginning of an article about empty introductions in marketing content, then deliberately gave it an empty introduction. It asked readers to think about the texts they enjoyed and described a world saturated with content. Yes, I made the draft do exactly what it was criticising.

The passage also included personal remarks, a joke, and an important qualification: a useful introduction need not always be short. Would the models remove the unnecessary opening? Keep the qualification? Preserve the humour? Or make the prose smoother while changing what I meant?

The draft, paragraph by paragraph:

  1. A generic opening about enjoyable writing and content overload.
  2. My observation about company content that takes too long to reach the point.
  3. Possible causes, including clumsy AI use, and why I close the tab.
  4. Scannable subheadings, with a joke about painful reading.
  5. A familiar corporate cliché as an example.
  6. What a lead needs to do, and why it can sometimes be longer.
  7. A promise of practical techniques, where the draft stops.
Read the full original draft

How do you write content people will actually read?

Start by thinking about the texts you enjoy reading yourself. Substantial, clear, structured, interesting from the first line, able to hold your attention, entertaining or informative. Ideally, both. The world around us is so saturated with content that capturing and holding attention has become quite a challenge. This is where tried-and-tested journalism and editing techniques can help.

When I read content on company websites, whether it’s blog posts, landing pages, core website pages, or case studies, I immediately notice when the opening has no substance. It starts with “once upon a time”, the history of the internet, or a very general description of problems everyone already knows about.

This can happen because of clumsy use of AI, or because the writer doesn’t understand how to edit these texts or how people read and consume information. As a result, potentially good material with valuable insights becomes boring from the first few lines. I’m patient, but even I’m quick to close the tab.

Sometimes a clear structure with subheadings you can scan quickly can save it. But even that is often a luxury. Then reading becomes painful, and nobody likes pain. Well, almost nobody.

My favourite is something like: “In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.” You’ve seen clichés like this, surely? They make my eyes bleed.

After the headline, the lead or introduction is the writer’s main opportunity to grab the reader’s attention and get them interested. And it doesn’t have to be super short. Sometimes even several paragraphs are justified if they’re needed to establish the topic and offer an important insight or a hook.

So how do you make a text interesting and readable? There are a few fairly simple techniques.

The two prompts

The General prompt specified the audience, requested clarity and concision, and asked for conversational language. It already gave the model a useful brief.

Read the General prompt

Please improve this draft for a blog aimed at marketers and business leaders. Make it clear, engaging, and easy to read. Improve the flow, remove unnecessary wording, and keep it conversational rather than corporate. Use British English.

Return the edited text and briefly explain the main changes.

Draft: [Insert the original draft here.]

The Editorial prompt added instructions about structure, what to preserve, how to cut, and how to avoid changing claims. It also asked the model to compare its edit with the original before returning it.

Read the Editorial prompt

Edit the draft below for a blog aimed at marketers and business leaders. Apply journalistic and editorial techniques to make the argument clearer and the writing tighter while preserving the author’s voice.

The material is about how structure and editing help writers get their point across. Edit the supplied passage, not an imagined full article. Do not add new sections or a conclusion to complete it.

Editorial instructions:

  1. Identify the main point. Check whether the opening communicates it or gives the reader a concrete reason to continue. Remove generic introductory material that adds nothing. Preserve a story or problem-led opening when it serves the piece; do not automatically force an inverted pyramid.

  2. Put the ideas in a logical order. Move paragraphs where necessary, remove duplication, and preserve the connections between ideas. Leave effective wording alone.

  3. Cut ruthlessly. Remove empty phrases, repetition, and explanations of the obvious. Prefer wording that says more with fewer words. Stop when another cut would remove useful meaning, context, a necessary qualification, humour, or the author’s voice. Do not replace repetition with synonyms and leave it in place.

  4. Keep the language conversational, direct, and specific. Preserve the author’s sharp phrasing and personal observations. Check for awkward phrasing, repetitive sentence patterns, and unnecessarily choppy sentences. Use natural transitions. Avoid corporate language, stock phrases, and rhetorical questions added just to create suspense.

  5. Preserve factual meaning and the strength of each claim. Do not turn “may” into “will”, an association into a cause, or an observation into a universal rule. Do not invent facts, statistics, quotations, sources, or the author’s experiences. Flag missing evidence rather than supplying it.

  6. Use British English and Oxford commas.

Before returning the edit, compare it with the original: check that you have not lost a substantive point, changed a claim, or added unnecessary explanation.

Return:

  • The edited passage.
  • A brief explanation of the main changes.
  • Any unresolved questions or evidence gaps, if needed.
<draft>
Insert the original draft here.
</draft>

In every run, the prompt included the full draft shown above. The placeholders here avoid repeating the same passage.

How I ran the experiment

I gave each model the same draft and both prompts, in separate chats. That produced six responses to compare.

The models received the draft and one of the two prompts, without my editorial guidelines or examples of other writing. This comparison covers those inputs. It doesn’t tell us how the same models would perform with a fuller brief or a library of reference material.

Item Setup
Date 4 November 2026
Material The beginning of an article, not a complete article.
Runs Three LLMs × two prompts = six responses.
Follow-up editing None in the saved test responses. Later work on the published article was a separate stage.
Evaluation My editorial assessment against the original draft and intended meaning, followed by an identity-masked review on 5 November 2026.

Models and settings

AI tool Model and setting Chat mode
ChatGPT ChatGPT 6, Medium Temporary / Unpersonalized
Claude Claude Opus 5.5 Incognito
Gemini Gemini 3.8 Thinking Temporary Chat

I used fresh temporary or incognito chats to reduce the influence of previous conversations and personalisation on the edits. The modes are listed above; I did not assume they made the three tools behave identically.

What I checked

I reviewed the edits for five things:

I also checked the models’ explanations against what they actually changed. These are my editorial judgements, not measurements of reader engagement. AI assisted with organising and analysing the responses; the initial assessment was not blind.

The results at a glance

The original passage contained 292 words, excluding its headline. All three Editorial responses were shorter. That didn’t make every cut a good one.

Model General prompt Editorial prompt
ChatGPT 338 words, +16%.
✅ Kept first-person observations and the qualification about longer introductions.
❌ Expanded some phrasing and changed the joke’s implication.
269 words, −8%.
✅ Opened with my observation and put the cliché beside the criticism.
⚠️ Moved the general paragraph further down, where I would still cut it.
Claude 313 words, +7%.
✅ Moved the example forward, kept the humour, and added signposts.
⚠️ Added “harder to pull off than ever”, an unsupported comparison.
256 words, −12%.
✅ Tightened the writing and retained much of my voice.
⚠️ Kept an opening I would remove and slightly strengthened the wording about possible causes.
Gemini 222 words, −24%.
✅ Put the cliché immediately after the criticism.
❌ Removed personal remarks and the subheadings point, narrowed the subject to B2B, and strengthened a causal claim.
206 words, −29%.
✅ Cut the generic opening and kept distinctive phrases.
⚠️ Broadened the closing claim and added an unsupported bounce-rate claim in its explanation.

✅ Keep · ❌ Reject · ⚠️ Question or revise. These are my judgements about particular edits, not scores for a whole model. Here are the changes behind them.

How I counted the words

I counted the edited passage, including internal subheadings, but excluded the main headline, output labels, and explanations. Contractions and hyphenated words count as single words; percentages are rounded. Gemini omitted the headline in its Editorial response, so I excluded headlines consistently. Missing spaces after punctuation in the saved documents do not merge words in this count.

What the General prompt changed

ChatGPT: more words, less of my phrasing

❌ Extra wording and a weaker joke.

The edit grew to 338 words. The generic opening remained, followed by another paragraph about competition for attention. Some additions changed the emphasis in ways I didn’t need:

Original: “I’m patient, but even I’m quick to close the tab.”

Edited: “And while I consider myself a patient reader, even I’m quick to close the tab.”

“Consider myself” adds a note of self-assessment. I wanted the direct statement, so I would keep “I’m patient”. The edit did retain my first-person perspective and the point that an introduction can be longer when it earns its length.

The change I found more revealing was in the joke:

Original: “Then reading becomes painful, and nobody likes pain. Well, almost nobody.”

Edited: “Reading becomes a chore, and nobody likes a chore. Well, almost nobody.”

The shape of the joke survived. The implication behind “almost nobody” didn’t. The model’s explanation said it had preserved my dry humour, which is why I check the edit rather than accepting its description of the edit.

Claude: a useful move, but the opening stayed

✅ Keep the example’s new position. ⚠️ Reconsider the opening.

Claude moved the corporate cliché next to the criticism of weak introductions. That made the example easier to connect to the point. It also added two subheadings and preserved the pain joke and the remark about clichés making my eyes bleed.

The passage grew by 21 words, 11 of them in the two subheadings. That extra length wasn’t all padding. Claude also explained “lead” as the journalistic term for an introduction, a useful addition for readers unfamiliar with it.

I would still cut the opening, which made a familiar general observation before reaching mine. I would also remove “harder to pull off than ever”. The draft offered no evidence for that historical comparison.

Gemini: shorter, but with a different voice and scope

❌ Cuts that changed the author’s voice and subject.

Gemini cut the passage to 222 words. It also replaced my personal account with broader claims about B2B writing, even though I had not limited the article to B2B.

Its explanation described one choice as:

“Replaced hyperbole with grounded authority”

That included removing “makes my eyes bleed”. The wording may not suit every publication, but it was a deliberate part of my voice. The General prompt didn’t explicitly ask the model to preserve every personal phrase; this result showed me why that instruction mattered for my intended piece.

The edit also removed the whole point about scannable subheadings. And “This can happen because…” became “This usually happens for two reasons…”, turning possible explanations into a claim about the usual causes. Those changes altered the argument as well as the tone. I would keep the earlier placement of the cliché, but reject these losses and the narrower B2B framing.

What the Editorial prompt changed

ChatGPT: the problem comes first, but the old opening returns

✅ An opening I would keep. ⚠️ The relocated paragraph still needs a decision.

ChatGPT began with my observation about company websites and followed it with the cliché. The result reached the problem sooner and preserved the pain joke and the personal observation about closing the tab.

I prefer that start because it shows the problem before discussing it. But the paragraph asking readers to think about the texts they enjoyed reappeared between the discussion of subheadings and the lead. It now offered a positive standard after the criticism, which is a defensible choice. For this piece, it still felt like a detour. I would cut it.

Claude: a reasoned defence of an opening I would remove

✅ Useful tightening. ⚠️ Check what stayed and what changed meaning.

Claude tightened the passage to 256 words but retained the opening. It explained:

“I kept your reader-centred opening, because it sets up the standard you go on to apply.”

That is an editorial argument I can assess. For this piece, I disagree: the later examples make the standard clear without that introductory list.

Another explanation deserved a closer look. Claude said “able to hold your attention” repeated “interesting from the first line”. Attracting attention and sustaining it are different jobs. But “capturing and holding” remained in the next sentence, so the idea hadn’t disappeared. The cut could be justified; that particular explanation didn’t justify it.

The edit also changed a qualified explanation:

Original: “This can happen because of clumsy use of AI, or because the writer doesn’t understand how to edit these texts or how people read and consume information.”

Edited: “Sometimes it’s clumsy use of AI. Sometimes the writer doesn’t understand how to edit or how people read and consume information.”

The second version states that these causes occur in some cases, rather than presenting them as possibilities. It is a slight increase in certainty; “sometimes” still limits the claim. I would retain the original qualification. Similar shifts appear in the General responses from Claude and ChatGPT, so this check applies to them too.

Gemini: useful cuts, with claims still to check

✅ An opening I would keep. ⚠️ Check the claims, including those in the explanation.

Gemini removed the generic opening, kept several distinctive phrases, and reduced the passage to 206 words. It moved the paragraph about scannable subheadings after the discussion of the lead. That gave me another ordering to consider.

Its final sentence was broader than my original promise of a few useful techniques:

“Writing content that people will actually read comes down to a few tried-and-tested journalism and editing techniques.”

I would soften that. These techniques help, but they do not explain everything that makes content worth reading.

There was also an unsupported claim in the explanation, rather than in the edited passage:

“In digital business writing, long leads often increase bounce rates”

No source accompanied that claim. That doesn’t establish that it is false, but I wouldn’t reuse it without checking. The request to explain how a longer introduction earns its length was useful. Even a model’s section about evidence gaps needs fact-checking.

What I would keep, question, and reject

Editorial decision What happened My response
Bring the example forward Five of the six responses placed the cliché next to the criticism; ChatGPT’s General response left it later. ✅ Keep. The example explains the problem immediately.
Remove the generic opening Gemini’s Editorial response deleted it. ✅ Keep the cut. The passage reaches my observation sooner.
Preserve the joke Both Claude responses and the Editorial responses from ChatGPT and Gemini retained “pain” and “Well, almost nobody.” ✅ Keep. The implication survives the edit.
Make the passage easier to scan Claude’s General response added two descriptive subheadings. ✅ Keep the idea. Review the headings when the full article is assembled.
Preserve the qualification about length All six edits retained the idea that an introduction can be longer when it has a purpose. ✅ Keep. Length alone was never the problem.
Keep or relocate the generic paragraph Claude’s Editorial response defended it; ChatGPT’s moved it down. ⚠️ Reconsider. I would still cut it from this piece.
Replace voice or change scope ChatGPT’s General response weakened the joke; Gemini’s removed personal remarks and introduced B2B. ❌ Reject those changes. They alter the voice or subject I chose.
Strengthen claims or add unsupported advice All three General responses introduced an unsupported historical comparison about attention. Gemini General asserted usual causes; Claude Editorial slightly strengthened possible causes; Gemini Editorial broadened the ending and added an unsourced bounce-rate claim in its explanation. ⚠️ Check the strength and support of claims under either prompt.

What the anonymised review changed

I had helped develop the Editorial prompt, and I preferred an opening that reached the problem quickly. Those preferences could shape my assessment. On 5 November, I asked ChatGPT, Claude, and Gemini to review the original draft and all six responses with the model names removed and the responses relabelled A–F. Each first assessed the edits without seeing my verdicts, then audited my earlier assessment. The General and Editorial prompt labels remained visible.

That check led to several changes:

The reviewers disagreed about paragraph order and made mistakes of their own. One said the models had not been matched across prompts, although each had received both. Another treated formatting that could have changed during copying as a model’s editorial decision. I checked those objections against the record before accepting them. The reviewers needed editing too. Apparently, there is no escape.

This was an additional editorial check with model identities hidden, not six independent blind verdicts: the second stage explicitly showed each reviewer the earlier assessment. It helped me refine the analysis, but it didn’t measure reader response or establish a winning prompt.

What I learned, and what I haven’t tested

For this draft, I found more edits I wanted to keep in the Editorial responses. But the General prompt also produced useful work, particularly Claude’s changes to structure and its preservation of humour. Detailed instructions did not prevent stronger claims or resolve what belonged in the opening. I would still edit every version before publishing it.

The experiment gave me specific things to check next time:

I ran each prompt once per model, without matched generation settings or reader testing. The prompts differed in several ways, so I cannot attribute an improvement to one instruction or establish a reliable advantage from these single responses. These results describe six edits of this draft; another draft or another run could turn out differently.

See the full responses

Each response is in a separate expandable block below. I’ve separated the edited passage from the model’s explanation and made the formatting easier to read. Paragraph spacing, heading presentation, and missing spaces after punctuation have been tidied up; the wording and substance have not been rewritten. The original saved responses remain the reference record.

ChatGPT — General prompt: full response

Edited passage

How do you write content people will actually read?

Think about the content you enjoy reading. It’s clear, well-structured and engaging from the first line. It has something worthwhile to say and holds your attention. It might be entertaining, informative or, ideally, both.

We’re surrounded by so much content that getting people’s attention — and keeping it — is harder than ever. That’s where tried-and-tested journalism and editing techniques can make a real difference.

When I read content on company websites — blog posts, landing pages, case studies or even core website pages — I often notice the same problem: the opening says almost nothing.

Instead of getting to the point, it starts with a lengthy backstory, a brief history of the internet or a generic description of a problem everyone already knows about.

Sometimes it’s the result of clumsy AI use. Other times, the writer simply hasn’t thought enough about editing or how people actually read and consume content.

Either way, potentially useful material with valuable insights becomes boring within the first few lines. And while I consider myself a patient reader, even I’m quick to close the tab.

A clear structure and a few well-placed subheadings can sometimes save the day. But often, even those are missing. Reading becomes a chore, and nobody likes a chore. Well, almost nobody.

One of my personal favourites is:

“In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.”

Sound familiar? Clichés like this make my eyes bleed.

After the headline, the opening is your best chance to give readers a reason to keep going. It needs to spark curiosity, offer something useful or make them feel they’ve come to the right place.

That doesn’t mean every introduction has to be short and punchy. Sometimes a few paragraphs are exactly what you need to set the scene, establish context and introduce an interesting idea. The key is to make every sentence earn its place.

So how do you make your content more interesting and easier to read?

A few simple techniques can go a long way.

Model’s explanation

Made the opening more direct. Replaced the list of adjectives with a more natural description of what makes content worth reading.

Improved the flow. Created clearer transitions between the problem of information overload, weak website copy, common causes and practical solutions.

Cut repetition and unnecessary wording. Tightened sentences without losing the original argument or examples.

Kept your personality. Preserved the conversational tone, first-person observations, dry humour and the memorable “make my eyes bleed” line.

Strengthened the key points. Gave the criticism of empty introductions more impact and clarified what a good opening should achieve.

Improved the transition to the next section. The ending now leads more naturally into the practical techniques that presumably follow.

One editorial choice worth highlighting: I kept the argument that introductions don’t always have to be short. It’s an important distinction: good editing isn’t simply about cutting words, but about making sure every sentence gives the reader a reason to continue.

ChatGPT — Editorial prompt: full response

Edited passage

How do you write content people will actually read?

When I read content on company websites, whether it’s blog posts, landing pages, core website pages, or case studies, I immediately notice when the opening has no substance. It starts with “once upon a time”, the history of the internet, or a general description of problems everyone already knows about.

My favourite is something like: “In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.” You’ve seen clichés like this, surely? They make my eyes bleed.

This can happen because of clumsy use of AI, or because the writer doesn’t understand how to edit or how people read and consume information. As a result, potentially valuable material becomes boring from the first few lines. I’m patient, but even I’m quick to close the tab.

Sometimes a clear structure with subheadings you can scan quickly can save it. But even that is often a luxury. Reading becomes painful, and nobody likes pain. Well, almost nobody.

Think about the texts you enjoy reading yourself. They’re substantial, clear, well-structured, and interesting from the first line. They hold your attention and entertain or inform. Ideally, both. With so much content competing for attention, tried-and-tested journalism and editing techniques can help.

After the headline, the opening is the writer’s main opportunity to capture the reader’s attention and give them a reason to keep reading. It doesn’t have to be super short. Even several paragraphs can be justified if they establish the topic and offer an important insight or a hook.

So how do you make a text interesting and readable? There are a few fairly simple techniques.

Model’s explanation

Moved the author’s observation to the opening. The original began with general advice about what makes content readable. The revised version starts with a recognisable problem on company websites, giving marketers and business leaders a more immediate reason to care.

Brought the cliché closer to the criticism. The “fast-paced digital world” example now follows the description of empty introductions. It demonstrates the problem rather than arriving several paragraphs later.

Removed repetition and tightened sentences. Cut overlapping references to capturing attention, unnecessary qualifiers, and wordy constructions. For example, “potentially good material with valuable insights becomes boring” becomes “potentially valuable material becomes boring”.

Improved the progression. The passage now moves from the problem, to a concrete example, to possible causes and consequences, then to what effective writing looks like and how editorial techniques can help.

Preserved the author’s voice. Kept “They make my eyes bleed”, the observation about closing the tab, “Well, almost nobody”, and the conversational questions. These give the piece personality without undermining its argument.

Kept the qualification about length. The point isn’t that all introductions should be short, but that they should earn the reader’s attention. The revised passage preserves that distinction.

Unresolved questions or evidence gaps

There are no factual claims that require evidence for this passage to work as a personal, observation-led introduction.

The suggestion that weak openings can result from clumsy AI use or poor editing remains an observation, not a proven causal relationship. The wording retains that qualification.

One editorial consideration: the passage promises “a few fairly simple techniques”. The material that follows should deliver specific, practical advice on structure and editing. Otherwise, the introduction risks making the same mistake it criticises: promising value without delivering it.

Claude — General prompt: full response

Edited passage

How do you write content people will actually read?

Think about the things you enjoy reading. They’re probably clear, well structured and interesting from the first line. They hold your attention because they either entertain or inform you, ideally both.

That’s harder to pull off than ever. We’re drowning in content, and grabbing someone’s attention (let alone keeping it) is a real challenge. This is where a few tried-and-tested techniques from journalism and editing can help.

The problem with most openings

When I read content on company websites, whether it’s blog posts, landing pages, core pages or case studies, I can tell straight away when an opening has nothing to say. It starts with “once upon a time”, a potted history of the internet, or a vague description of problems everyone already knows about.

My personal favourite goes something like this: “In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.” You’ve seen it. We all have. It makes my eyes bleed.

Sometimes this is down to clumsy use of AI. Sometimes the writer simply doesn’t know how to edit, or how people actually read online. Either way, the result is the same: material that might contain genuinely useful insights turns dull within a few lines. I’m a patient reader, but even I close the tab.

A clear structure with subheadings you can scan can sometimes rescue a piece. Often, though, even that’s missing. Reading becomes painful, and nobody likes pain. Well, almost nobody.

Make your introduction earn its place

After the headline, your introduction (or “lead”, as journalists call it) is your best chance to hook the reader. It doesn’t have to be short. Sometimes several paragraphs are justified, as long as they set up the topic and offer a real insight or a reason to keep reading.

So how do you make your writing interesting and readable? Here are a few simple techniques.

Model’s explanation

Tighter opening. I replaced the list of adjectives (“substantial, clear, structured…”) with two short sentences, so the piece starts with energy rather than a catalogue.

Reordered for flow. The cliché example now sits right after you describe weak openings. It’s the perfect illustration of that point, and in the original it arrived two paragraphs later and felt detached.

Cut wordiness. For example, “The world around us is so saturated with content that capturing and holding attention has become quite a challenge” became “We’re drowning in content…”, and “how people read and consume information” became “how people actually read online”.

Added subheadings. The piece argues for scannable structure, so it helps if it practises what it preaches. Drop them if your blog format doesn’t suit them.

Kept your voice. The humour (“nobody likes pain. Well, almost nobody”, “makes my eyes bleed”) stays, with a little extra rhythm (“You’ve seen it. We all have.”).

British English and conventions. “Favourite” is kept, and the phrasing follows UK usage (“potted history”, “straight away”). If your house style uses single quotation marks, swap the doubles.

Claude — Editorial prompt: full response

Edited passage

How do you write content people will actually read?

Think about the texts you enjoy reading: substantial, clear, structured, and interesting from the first line. Entertaining or informative. Ideally, both. With so much content competing for attention, capturing and holding it has become a real challenge. This is where tried-and-tested journalism and editing techniques can help.

When I read content on company websites, whether blog posts, landing pages, core pages, or case studies, I notice straight away when the opening has no substance. It starts with “once upon a time”, the history of the internet, or a general description of problems everyone already knows about.

My favourite is something like: “In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.” You’ve seen clichés like this, surely? They make my eyes bleed.

Sometimes it’s clumsy use of AI. Sometimes the writer doesn’t understand how to edit or how people read and consume information. Either way, potentially good material with valuable insights becomes boring within the first few lines. I’m patient, but even I’m quick to close the tab.

A clear structure with scannable subheadings can sometimes save it, but even that is often a luxury. Then reading becomes painful, and nobody likes pain. Well, almost nobody.

After the headline, the lead is the writer’s main opportunity to grab the reader’s attention. It doesn’t have to be super short. Several paragraphs can be justified if they’re needed to establish the topic and offer an important insight or hook.

A few fairly simple techniques can make a text interesting and readable.

Model’s explanation

Opening: I kept your reader-centred opening, because it sets up the standard you go on to apply. I trimmed the list of qualities: “able to hold your attention” repeated “interesting from the first line”. I also replaced “The world around us is so saturated with content” with a shorter clause that makes the same point.

Order: I moved the “In today’s fast-paced digital world” example up so it sits directly after your description of weak openings. That way the reader gets the concrete example before the explanation of why it happens. Previously it appeared after the subheadings point, which broke the thread.

Causes paragraph: I split “This can happen because of… or because…” into two short parallel sentences and added “Either way” to link them to the consequence.

Cuts: I removed “yourself”, “very”, “or introduction”, “and get them interested” (already covered by “grab the reader’s attention”), and “And” before “it doesn’t have to be”. I changed “from the first few lines” to “within”, which reads more naturally with “becomes boring”.

Final line: I turned the closing rhetorical question into a statement so the passage leads straight into the techniques without a suspense question. If you prefer the question as a deliberate signpost, it works fine to restore it.

Voice preserved: I kept “make my eyes bleed”, “I’m patient, but even I’m quick to close the tab”, “Well, almost nobody”, “super short”, and your rhetorical “surely?”. These carry your personality.

Open questions

Headline or first line? I’ve treated “How do you write content people will actually read?” as the headline. If it’s meant to be the first sentence of the body, consider replacing it with something more specific, since the piece itself warns against generic openings.

“Once upon a time”: Is this literal, or shorthand for long-winded backstory? If it’s shorthand, a real example like your cliché quote would land harder.

AI claim: Blaming clumsy AI use is presented as your observation, which is fine. If you want it to carry more weight with business readers, an example or source would help.

Ending: The passage ends by promising techniques it doesn’t list. I haven’t added them, as you asked. Just make sure the next section follows immediately.

Gemini — General prompt: full response

Edited passage

How do you write content people will actually read?

Start by thinking about what holds your own attention. It is usually writing that has substance, gets straight to the point, and offers genuine insight—ideally entertaining, informative, or both. In a market flooded with noise, earning a reader’s focus has never been harder. This is where classical journalism and sharp editorial instincts make all the difference.

Yet far too much B2B copy—from blog posts to case studies—falls flat before it even begins. You know the type: an opening that recites the history of the internet, waffles through obvious background, or serves up a lifeless corporate cliché:

“In today’s fast-paced digital world, businesses must adapt to stay competitive.”

We’ve all seen it. It’s filler, and it’s an immediate signal to close the tab.

This usually happens for two reasons: clumsy AI prompting, or a writer who hasn’t considered how busy professionals actually consume information. Brilliant insights get buried beneath throat-clearing prose, and expecting a reader to wade through fluff is wishful thinking.

Your headline earns the click, but your opening paragraph earns the read. It doesn’t have to be artificially brief—a longer setup works if every sentence builds momentum and delivers value—but it must earn its keep.

So, how do you hook an audience and keep them reading? Here are a few straightforward techniques to sharpen your copy.

Model’s explanation

Tightened the pace and eliminated filler: Removed redundant phrasing (such as “the world around us is so saturated” and “texts you enjoy reading yourself”) to reach the core argument much faster.

Separated the cliché for impact: Pulled out the generic “fast-paced digital world” quote into a blockquote so it immediately resonates as a cautionary example.

Replaced hyperbole with grounded authority: Swapped overly dramatic phrasing like “makes my eyes bleed” with punchy, peer-level observations that feel natural rather than exaggerated.

Refined the messaging on introductions: Articulated the editorial function of the lead clearly (“Your headline earns the click, but your opening paragraph earns the read”), making the advice more actionable for marketing and business leaders.

Unified tone and British spelling: Adjusted sentence cadence to maintain an engaging, editorial voice that avoids both corporate jargon and casual rambling.

Gemini — Editorial prompt: full response

Edited passage

When I read content on company websites—whether blog posts, landing pages, or case studies—I immediately notice when an opening lacks substance. It starts with ‘once upon a time’, the history of the internet, or a generic description of problems everyone already knows about.

My favourite: ‘In today’s fast-paced digital world, technology is evolving rapidly, and businesses need to adapt to stay competitive.’ You’ve seen clichés like this, surely? They make my eyes bleed.

This can happen because of clumsy use of AI, or because the writer doesn’t understand how to edit or how people consume information. As a result, potentially valuable material becomes boring from the first few lines. I’m patient, but even I’m quick to close the tab.

After the headline, the lead is the writer’s main opportunity to capture the reader’s interest. It does not have to be brief; several paragraphs are justified if they establish the topic and offer an important insight or hook.

A clear structure with scannable subheadings can sometimes save a piece, but even that is often a luxury. Without it, reading becomes painful, and nobody likes pain. Well, almost nobody.

Writing content that people will actually read comes down to a few tried-and-tested journalism and editing techniques.

Model’s explanation

Eliminated Generic Throat-Clearing: Removed the opening rhetorical question, the prompt to “think about texts you enjoy”, and the cliché about a “content-saturated world”. These added no new information and delayed the argument. Opening immediately with the author’s critique gives the reader a concrete problem to engage with.

Reordered Ideas for Narrative Logic:

Moved the cliché example (“In today’s fast-paced digital world…”) directly below the description of hollow openings, rather than leaving it isolated after the discussion of structure.

Grouped the critique of the lead, the root causes (AI and lack of editing), and the editorial standard for leads together, before transitioning into body structure (subheadings). This follows the reader’s natural journey down a page: headline lead structural layout.

Pruned Repetition and Wordiness:

Condensed “core website pages, landing pages…” to remove overlapping site categories.

Removed tautologies such as “grab the reader’s attention and get them interested” and “read and consume”.

Converted “subheadings you can scan quickly” to “scannable subheadings”.

Preserved Voice and Conversational Tone: Kept personal, characterful touches intact, including the opening observation, the quip “They make my eyes bleed”, and the dry aside “Well, almost nobody.”

Applied British English and Oxford Commas: Standardised spelling (favourite), applied serial commas across lists, and used single quotation marks for referenced phrases.

Unresolved Questions & Evidence Gaps

The AI Claim: The draft asserts that weak openings can result from the “clumsy use of AI”. If this piece targets marketers and business leaders, briefly specifying what clumsy AI usage looks like here (e. g. unprompted corporate boilerplates or raw generic summaries) would strengthen the critique.

The Multi-Paragraph Lead: The author defends multi-paragraph leads if they deliver “an important insight or hook”. In digital business writing, long leads often increase bounce rates; clarifying how a longer lead maintains reader momentum (or providing a brief example) would support this claim.

Try it with your own writing

Use a draft where you know what you meant and which details matter to you. Give each prompt to a fresh chat, save the first outputs unchanged, and record the date, model, and settings. Compare what moved, disappeared, or was added before reading the model’s explanation of its changes.

If you repeat the test, keep those responses too. The edit that challenges your conclusion may be the most useful one to inspect.

For the journalism and editing methods I used to finish the article, read Writing with AI? Think like a journalist. Give people a reason to read.

Read the main article →

Spotted something I missed, disagree with a verdict, or tried this yourself? I’d like to hear it.

Send feedback →

I read every message and do my best to reply within three days. I’ll only quote you with your permission.


About the language versions: I also ran this experiment in Ukrainian, with six responses per language. Each version presents its own results. I adapted and edited the source texts for their language and cultural context; the two versions cover the same subject without being word-for-word translations or unedited machine translations. They are separate editorial adaptations, not a controlled comparison of language performance.

Agree or disagree?

Tell me what you think. Point out what I missed, challenge an argument, or suggest a topic.

I’ll use your details only to read and reply to your message. See the Privacy notice for how I handle your data.

Let’s talk

Describe the problem, and I’ll tell you straight whether I can help.

I’ll use your details only to reply to your enquiry. See the Privacy notice for how I handle your data.