10 Inventive Methods to Write with AI (With out Shedding Your Soul)

So, after years of listening to the identical message, folks began associating AI-generated content material with low-effort, mass-produced slop. AI-assisted content material earned a nasty fame earlier than it had an opportunity to mature.

This text is my try to reset the dialog.

I’ll share how we use AI at Ahrefs to create content material, together with some content material experiments we’ve been working. To not substitute human considering, however to make doable issues that was once too troublesome, too costly, or just not possible.

My purpose isn’t to persuade you to automate extra. It’s that can assist you see AI as a artistic device somewhat than a content material manufacturing facility.

And another factor: I feel you’ll really get pleasure from most of those concepts. Individuals typically say AI makes artistic work much less enjoyable. Used nicely, I feel it might do the alternative.

There are occasions when you already know what you wish to say, however you don’t wish to agonize over each sentence. That’s the place vibewriting comes in.

Vibewriting is steering AI with tough inputs and iterative suggestions somewhat than making an attempt to engineer the proper immediate or get a completed piece in a single shot. You give it context, react to what it produces, and steadily form the output till it matches what you need.

Begin by letting AI produce a primary draft, then deal with it like one thing to edit, not one thing to publish. Ask it to make the writing punchier, reduce the introduction, increase a piece, tighten a paragraph, or rewrite a weak transition. Every spherical of suggestions strikes the draft nearer to what you had in thoughts.

Instance

I used this methodology to put in writing Agent-To-Agent Advertising Was Simply Born on Moltbook. I requested Letaido (AI marketing platform by Ahrefs) to get some knowledge on Moltbook.com, gave it some notes from my guide analysis, and an arc of the story that I already had in my head, and requested it to mix every thing collectively in an article.

Our Director of Content material Advertising, Ryan Law, tried out this methodology and stated: “it was probably the most enjoyable I’ve had writing for Ahrefs in ages.” Try his video:

Vibewriting additionally works with different sorts of content material, like presentation decks. Right here’s one I made for a webinar. You may take a look at the complete interactive deck
here, and right here’s the webinar the place I used it.

Beginning immediate

I wish to vibewrite a weblog publish about [topic]. This is my basic concept for the article [describe the idea]. I've gathered these supplies up to now [attach anything you'd like the AI to use and reference] and right here is the kind of article I am after https://ahrefs.com/weblog/creative-ways-to-write-with-ai/. Let's begin with the summary of the article and the define.

Strive with:

• Newsletters
• Opinion items
• Essays
• Quick analysis items

You’re circling a subject that received’t sit nonetheless. Concepts preserve touchdown at odd hours: a hyperlink a colleague sends, a screenshot, a thought in your commute you don’t wish to lose. None of it’s prepared for a top level view but, and forcing construction this early would kill it.

So don’t. Hold one draft completely open and throw every thing into it. Each time you add one thing, AI folds it in, and the piece thickens. Nothing is ever “began” or “completed.” It’s simply the perfect present synthesis of what you’ve collected. Construct that pile as soon as, and you’ll render it as an article right this moment, a chat subsequent month, three posts after that.

I name it the Residing Draft methodology. It’s a bit much like vibewriting. The distinction is that with vibe writing, you steer a draft towards a vacation spot you have already got in thoughts, and with the residing draft, you don’t have a vacation spot but—you feed a subject over time and let the vacation spot reveal itself.

Instance

I wanted this workflow so badly that I ended up constructing a customized app for it with Letaido.

I’ve been utilizing it to doc every thing that’s occurred in AI notion optimization since I printed my experiment in December 2025: follow-up experiments, commentary, real-world case research, and milestone occasions—like Google being sued over the accuracy of its AI Overviews, to place it politely.

I begin with a working title and an issue assertion.

After which I simply drop in no matter materials I discover and watch AI unfold the story.

Beginning immediate:

Deal with this chat as a residing draft. At any time when I add new materials, combine it naturally into the article, take away repetition, enhance the construction, and level out gaps or contradictions with out rewriting my concepts.

And if you need an app like mine, present this GitHub repo to your AI agent: https://github.com/mmakosiewicz/self-building-articles-app 

Strive with:

• Analysis-heavy articles
• Lengthy-term writing initiatives
• Matters you’re nonetheless exploring

Think about you wish to write about one thing you already know inside out. You’ve accomplished the work, realized the teachings, and have insights you genuinely assume are price sharing.

Now the onerous half: turning every thing in your head into one thing that’s clear and interesting for people who find themselves ranging from scratch.

That’s the place AI can assist.

As an alternative of asking it to put in writing the article, ask it to interview you. Let it ask considerate questions, reply them as for those who’re speaking to a different individual, and use these solutions as the muse for the piece.

Instance

I’m utilizing this methodology to put in writing up an web optimization experiment on whether or not a structured FAQ may help AI assistants retrieve correct details about Ahrefs.

What I discover most helpful is that it helps me escape the curse of data. As a result of the AI doesn’t share all of the context that’s already in my head, it naturally exposes the gaps in my considering and forces me to clarify concepts extra clearly. The result’s normally a greater article than if I’d tried to put in writing it from reminiscence alone.

Beginning immediate

Interview me for an article about [topic]. Ask one query at a time like an skilled journalist. Problem obscure solutions, ask for examples, and preserve digging till you've gotten sufficient materials. Then flip the dialog into a refined article whereas preserving my voice.

Strive with:

• Thought management
• Founder tales
• Case research and experiments
• Opinion items
• Classes realized

I’ve seen that many questions don’t really need new solutions. Whether or not folks phrase them in another way or ask from a barely completely different angle, the underlying reply is commonly the identical. And as a rule, we’ve already written it someplace on our weblog.

The problem is extra about discovering the correct items and presenting them in a means that matches the query somewhat than creating new information

So after I run into this example, I level AI at our source-of-truth paperwork and let it do the digging. It finds the related passages, removes duplicate concepts, and assembles a draft that’s grounded in what we already know.

Instance

At the least 70% of this text is “recycled” from info we’ve already printed. We already had every thing we wished to say about AI chatbot visitors—it was simply scattered throughout dozens of weblog posts. So, as a substitute of writing it from scratch, I guided AI to drag these items collectively right into a coherent article.

In case you ask me, it turned out fairly nicely. It genuinely helps you perceive AI chatbot visitors, reveals you the way to observe it, and it even ranks.

Higher but, it launched a unique search intent into the highest 10. That’s simpler to drag off with low-KD key phrases, I do know—however I’ll take it.

The one motive I may put this text collectively so shortly was that I’d already constructed the infrastructure behind it: a “supply of reality” repository containing product documentation, Ahrefs how-tos, insights from our knowledge research, and different key assets.

At any time when I come throughout an necessary inner web page, I add its URL to the app. It distills the important thing info and syncs it on GitHub, so later I can merely ask, “What do the SoTs say about this?” and immediately pull the related context right into a draft.

Beginning immediate

Search my documentation for every thing associated to [topic]. Pull collectively probably the most related info, establish recurring themes, take away overlap, and draft an article that builds on present information as a substitute of inventing new content material.

And if you need an SOT app like mine, present this hyperlink to your AI agent:
https://github.com/mmakosiewicz/sots_webinar

Strive with:

• Product explainers
• Evergreen articles
• Documentation
• Guides and how-tos
• Updating outdated content material

Among the finest content material begins with knowledge.

In these instances, the phrases are simply there to clarify what the numbers reveal. And likelihood is, you have already got helpful knowledge sitting inside your online business: product utilization, buyer conduct, marketing campaign efficiency, experiments, surveys, help tickets, or gross sales information.

The problem is discovering the tales hidden inside it. That’s the place AI shines.

Feed AI the info and ask it to research. Have it search for outliers, sudden patterns, shocking correlations, or questions price exploring. Then construct the article across the insights that emerge.

Instance

In case you’d prefer to see what data-driven content material seems like in follow, listed below are just a few latest examples written by Ryan Regulation and Louise Linehan.

We constructed these with Letaido, which has been an enormous unlock for working with Ahrefs knowledge. In contrast with a normal MCP setup, it provides us entry to extra knowledge endpoints, can work autonomously, and comes with native integrations like WordPress, so we are able to publish content material immediately from the device.

Letaido dealt with the heavy lifting: connecting to Ahrefs knowledge, calling APIs for specialised databases, producing visualizations, and even serving to write elements of the articles.

Si Quan from our content material workforce even constructed a customized Letaido app to automate the method of updating data-driven articles like these.

As an alternative of rebuilding every article from scratch each time the info adjustments, the app refreshes the numbers and generates an up to date draft, making it a lot sooner to maintain our analysis present.

On this information, he explains how he constructed it, walks by the complete course of, and reveals the way it sends an electronic mail notification when new knowledge is able to evaluation—so you’ll be able to comply with the identical method your self.

Beginning immediate:

I am attaching a dataset from our enterprise. Do not write an article but.
First, analyze the info like an investigative journalist or analyst. Search for:
- shocking patterns or outliers
- tendencies over time
- correlations price exploring (do not assume causation)
- rankings and benchmarks
- something that contradicts widespread assumptions
- questions the info raises
- findings that might make a robust headline As soon as you've got analyzed it, suggest 10 article concepts primarily based on probably the most fascinating discoveries. For every one, clarify why it is fascinating and what further evaluation (if any) would strengthen the story.

Strive with:

• Unique analysis
• web optimization research
• Business studies
• Product insights
• Knowledge journalism
• Instance

In 2026, an OpenAI model solved a geometry problem that had stumped mathematicians for 80 years. The breakthrough was that it explored an method people had dismissed. Researchers spent many years making an attempt to show the accepted reply as a substitute of following an unpromising path. The AI had no such bias or impatience, so it discovered what everybody else missed.

Brainstorming works the identical means. Most individuals cease after their first few first rate concepts—the identical apparent ones everybody else has. AI retains going.

You may actually ask AI for “100 methods to consider this,” then cluster the concepts or increase the perfect ones. It’ll floor angles you most likely wouldn’t have thought of. Your job is deciding which of them are price pursuing.

Instance

My colleague Si Quan informed me about this methodology, and I’ve at all times been impressed by the titles and angles he comes up with. So I made a decision to strive it with an concept that retains coming again to me each time I analysis AI web optimization: model is content material.

It surfaced just a few angles I’d already explored, which gave me confidence it was heading in the right direction. However it additionally uncovered a number of concepts I’d by no means thought of.

Listed below are a number of the new views I found due to this method:

By the best way, this methodology is an efficient instance of how AI can increase your work, not solely automate it.

Beginning immediate

Give me 100 methods to consider [topic with a brief explanation of how you interpret it]. Cluster comparable concepts.

Strive with:

• Brainstorming angles and matters ought to work with any kind of content material.
• May very well be a superb method for repurposing longer content material items for social media short-form content material.

One among AI’s largest strengths is how adaptable it’s; possibly much more than people. You may ask it to assume in a selected means, and it’ll change approaches immediately.

You should utilize that to your benefit in content material advertising. As an alternative of asking AI to generate concepts from scratch, give it a confirmed considering framework to work inside.

framework provides the mannequin a transparent path to comply with, challenges weak assumptions, and helps produce articles that specify, diagnose, or argue—not simply summarize.

So somewhat than prompting it to “write an article about [topic],” begin by giving it a solution to assume: Jobs to Be Finished, the Concept of Constraints, Porter’s 5 Forces, a choice tree, first rules, and even your individual mental model.

Instance

That is one other method my colleague Si Quan launched me to. I already knew you may ask AI to tackle a task—like an information analyst, a lawyer, or a troublesome editor—however this method felt extra structured and managed. So, let’s strive it in Letaido utilizing Opus 4.8.

The outcome was an in depth report with your entire reasoning course of specified by entrance of me. Two sections stood out particularly.

The primary was the place the AI challenged its personal conclusions, questioned its assumptions, and labored its means towards what it thought of the strongest rationalization.

The second was seeing these insights make their means into the article itself. It wasn’t simply reasoning for reasoning’s sake—the AI really carried its conclusions by into the ultimate draft.

I don’t know whether or not the AI genuinely reasoned its means by the issue or just simulated the method. And it positively didn’t produce one thing I may publish as is.

However that wasn’t the level.

It bought me a lot additional than a clean web page would have, and it helped me manage my very own considering.

That’s extremely helpful as a result of good writing begins with good considering—and considering continues to be the onerous half. It’s not one thing we are able to absolutely outsource to AI.

Beginning immediate

Use the Concept of Constraints Logical Pondering Course of to investigate
[topic]. First, construct the suitable logic tree for any such
article. Establish the seen signs, root causes, assumptions, constraints, and certain results of the proposed answer. Problem weak causal hyperlinks earlier than writing. As soon as the tree is sound, flip it into a transparent article with a robust argument.

Strive with:

• Opinion items
• Product decision-making guides

Some articles don’t want a recent burst of creativity. They should come out the identical means each time. Launch notes, recurring roundups, touchdown pages: you already know the method. A single mega-prompt making an attempt to do it abruptly provides you inconsistent high quality you’ll be able to’t belief throughout a workforce.

Break it right into a pipeline as a substitute; a set of AI expertise chained collectively. Analysis, sources, temporary, define, draft, confirm, format, with a pause to your sign-off at every gate. AI does the phases between. You approve on the checkpoints, so errors get caught early as a substitute of compounding.

How is that completely different from typical AI content material automation?

• The workflow follows your confirmed course of. It isn’t inventing a brand new means of working every time, which makes the output extra in line with the way you already write.
• You management the inputs and keep concerned all through. Since you’re invested in every stage, it’s a lot simpler to guage the standard, spot issues, and enhance the system over time.
• It’s comparatively fast to create and simple to vary. That’s as a result of the workflow is constructed from particular person AI expertise somewhat than locked inside a closed-source device. You don’t want deep technical information or pages of documentation to regulate it, both.
• It will also be extra resilient than a inflexible automation. If one step fails, the AI can typically diagnose the issue, revise the instruction, or strive a unique method as a substitute of merely stopping the workflow (not like an n8n automation).

Instance

Ryan Regulation constructed an app like this utilizing Letaido. You give it a subject and some supply hyperlinks, and it takes care of the remainder. It researches the subject, creates an editorial temporary, builds a top level view, writes the article, fact-checks each declare, and pauses at three key phases so you’ll be able to evaluation and approve the route earlier than it strikes on.

Right here’s Ryan explaining the app:

Beginning immediate

Construct me an assisted long-form article pipeline. Atomic enter is a goal key phrase. Levels run sequentially as background jobs the UI polls: (1) key phrase analysis by way of Ahrefs, (2) competitor SERP fetch, (3) AI Content material Helper subject snapshot, (4) bulleted define with mandated subject protection, (5) data-mention placement, (6) full draft, (7) polish, (8) WordPress shortcode formatting + .docx export. Every stage reveals its output, has an "edit" textarea, and a "refine with suggestions" chat that re-runs the stage with my notes. Type information comes from a per-author voice profile.

Strive with:

• Recurring weblog posts
• Product bulletins
• Documentation
• Touchdown pages
• Editorial workflows

Buyer conversations have at all times been top-of-the-line sources of article concepts. They include actual questions, requested in your clients’ personal phrases, and you’ll even see which of them come up most typically.

The issue was that uncovering these insights meant manually studying by hundreds of help tickets, chat logs, and gross sales name transcripts. The data was at all times there—it simply wasn’t sensible to entry at that scale.

That’s what AI adjustments.

Level it at these conversations, and it might analyze all of them, group comparable questions collectively, evaluate them in opposition to your present content material to keep away from duplicates, and establish the gaps in your content material library. The questions your clients ask most frequently turn out to be the guides they’re really in search of.

Instance

This methodology works with any form of buyer help/CRM product so long as it presents an API or MPC with entry to buyer conversations. On this instance, I’ll be utilizing Fin (Intercom) with Letaido dealing with the MCP.

I discovered some untapped matters with only a few minutes of working with the info. Apparently, some customers had hassle discovering inner hyperlink knowledge and skilled points fetching knowledge with Google Knowledge Studio.

AI was even capable of generate some first rate solutions to those questions:

Kudos to Kamila Olexa for the concept!

Strive with:

• Assist heart articles
• Product documentation
• FAQs
• Buyer schooling
• Backside-of-funnel content material

Beginning immediate

Earlier than we begin, right here’s one tip for utilizing AI to investigate knowledge: don’t ask it to interpret knowledge you haven’t checked out your self. As an alternative of asking for the ultimate reply instantly, ask AI to indicate you the obtainable knowledge first and clarify what it’s seeing.

AI can nonetheless hallucinate or take shortcuts, particularly when analyzing massive datasets. For instance, we had round 7,500 Intercom conversations in a single month—far an excessive amount of to investigate reliably in a single go.

Right here’s a immediate to start out that form of evaluation:

I wish to establish gaps in our documentation, however do not generate suggestions but.
First, analyze our buyer conversations and present me the info.
Please:
- Group comparable buyer questions into themes.
- Depend how typically every theme seems.
- Embody consultant examples from actual conversations.
- Present the precise wording clients use each time doable.
- Flag any uncertainty or themes which will overlap. Don't recommend new articles but. I wish to evaluation the grouped questions earlier than we determine what to doc.

After reviewing the output, you’ll be able to comply with up with:

Now evaluate these themes with our present assist heart and documentation.
For every theme:
- Inform me whether or not it is already lined.
- Level to the present article if one exists.
- Establish lacking or outdated content material.
- Rank the gaps by how typically clients ask about them. Then recommend the highest 10 documentation alternatives, explaining why every one deserves to exist.

A extra dependable method is to have AI monitor new conversations as they arrive in as a substitute of asking it to dig by months of historic knowledge abruptly. Breaking the duty into smaller, ongoing analyses is each simpler for the AI and far much less prone to produce deceptive outcomes.

Any further, monitor new buyer conversations as a substitute of analyzing your entire historical past each time.
At any time when new conversations can be found:
- Group recurring questions into themes.
- Spotlight any new matters that have not appeared earlier than.
- Monitor which questions have gotten extra widespread.
- Evaluate new questions in opposition to our present documentation.
- Alert me when a recurring query is not answered by our assist heart. For each advice, embody:
- What number of conversations point out it.
- Instance buyer messages.
- Associated documentation (if any).
- A instructed article title and a brief define. By no means assume conclusions with out exhibiting the supporting dialog knowledge first.

Documentation begins going old-fashioned the second you ship the subsequent launch. A setting will get renamed, a restrict adjustments, a brand new function launches, and all of the sudden, a assist article is not correct.

The identical is true for product advertising content material like purchaser’s guides and comparability pages. In lots of instances, it’s even more durable to maintain these updated as a result of you need to observe adjustments in each your individual product and your opponents’.

That’s an issue for each web optimization and person expertise.

Happily, AI can deal with a lot of that work. All it wants is a listing of the pages you wish to preserve, the sources the place it ought to search for updates, and—for those who select to present it entry—your CMS, so it might replace every thing mechanically.

Instance

My colleague Kamila Olexa constructed a system like that utilizing Claude Code and Firehose. Firehose (by Ahrefs) is a real-time net knowledge streaming API that constantly displays adjustments throughout the general public net and pushes matching updates to your software as they occur.

The workflow is constructed round automation with a human approval step. In a nutshell:

1. Firehose constantly displays your opponents’ pricing pages and triggers the workflow each time certainly one of them adjustments.
2. Claude then extracts the up to date pricing into structured knowledge, identifies which of your articles point out that competitor, and rewrites solely the affected sections as a substitute of your entire publish.
3. Moderately than publishing mechanically, the workflow sends a abstract of the proposed adjustments to Slack, the place you’ll be able to shortly evaluation what shall be up to date.
4. A easy ✅ response approves the edits, after which the workflow updates the related pages in your CMS and publishes them mechanically.

Beginning immediate

As an alternative of a beginning immediate, I’ll leave you with Kamila’s article. It explains her workflow from begin to end, so you’ll be able to copy the identical method your self.

Strive with:

• Product documentation
• API documentation
• Assist facilities
• Inside information bases
• Launch notes
• Characteristic comparability pages
• Authorized or coverage adjustments

AI bros found a planet fabricated from gold and determined the perfect use for it’s mass-producing low-cost jewellery. You may have a greater choice.

You should utilize AI to make higher content material whereas having fun with the method. The catch is that you need to keep concerned. The extra you contribute, the higher the result. I feel that’s the course correction we have to make with AI.

Thanks for studying! Come and say hello on LinkedIn or Substack.