When our weblog group made the change to an AI-first writing mannequin in June 2025, my response was cut up. There was pleasure on one facet: I might see the inefficiencies in our course of that AI would possibly really repair. On the opposite, nervousness. And whereas loads of that nervousness was concerning the expertise itself (how nicely these instruments might really write, how a lot we might belief them), my major concern was my group.
I’ve managed HubSpot’s weblog writing group for 3 years, and earlier than that was a HubSpot Weblog author myself. I perceive what it means to have possession over your course of. Most of the writers I handle bought into this work as a result of they love writing. They benefit from the craft of discovering the appropriate sentence and incomes a reader’s consideration. Site visitors numbers and conversions are secondary. Throughout this AI transition, I felt the burden of what it might imply for these whose whole strategy to their craft was altering quick.
A lot of the general public dialog about AI content material will get the story mistaken. There’s an assumption that as a result of AI can write passably nicely, the exhausting half is over. Simply apply the identical low-effort polish you’d apply to an AI-drafted electronic mail or memo to your weblog put up, then ship it.
However I can inform you from expertise: it doesn’t work that method. A weblog put up is supposed to compete towards dozens (or a whole bunch) of different posts on the identical subject. On this sport, generic, low-effort writing won’t win, and that’s excellent news on your content material group. It signifies that though the writing course of has modified, the craft stays essential.
So, how do you steadiness craftsmanship with the effectivity beneficial properties AI guarantees? And the way do you construct a course of that takes benefit of what AI is definitely good at with out falling into its traps? We spent the final yr figuring that out. Right here, I’ll share what we discovered: the place time was saved, the place it bought added, and what it takes to make AI work with out compromising the standard requirements you’ve spent years constructing.
Desk of Contents
Earlier than AI, writing was private.
Earlier than AI, our writing course of was, nicely, a writing course of. Whereas we had editorial requirements and guardrails to make sure constant high quality, we didn’t implement a particular technique for a way posts needs to be written. We employed writers and editors for his or her capacity to craft, and we trusted them to make use of that capacity. We understood that writing is greater than only a technical talent. Giving our writers the area to develop their very own particular person workflows yielded content material with character and distinctive insights.
When coaching new writers, we taught a easy, versatile framework to assist people begin to construction their time:
- Outlining and planning took round half-hour to an hour.
- Analysis might take one to 2 hours, relying on the complexity of the subject and the author’s familiarity with it. This included sourcing quotes, discovering unique knowledge, and writing from private expertise.
- Drafting, the half the place somebody really places phrases to the web page, took round 5 to seven hours on common for a primary draft.
- Modifying, QA, formatting, and finalizing in our CMS added one other roughly two hours on high of that.
For the standard weblog put up, you’re simply a full day’s work, often longer. The majority of that point lived within the drafting part — writing the phrases, discovering the appropriate construction, and determining the way to clarify one thing clearly was the core of the job.
What We Realized From Letting Writers Lead
Once we formally adopted AI-assisted writing in June 2025, my strategy was deliberate: don’t over-prescribe. Permitting our writers to develop their very own processes had served us nicely for over a decade, and we needed to take care of that sense of management for our writing group as a lot as potential, at the same time as we added AI to the combo.
What followed was not the clean efficiency win I expected. AI solved some problems and introduced new ones. But by working through those problems with our writers, we built a process that we know works.
Faster drafting doesn’t mean faster content.
At first, the signs were good. Writers were putting words on the page faster than before, and I assumed that would translate directly into faster posts overall.
It didn’t.
A couple of months in, after reviewing time tracking data and having conversations with my team, a more complicated picture emerged. The quantity of posts stayed about the same, but posts were getting longer, more detailed, and more demanding to produce. Writers were spending more time testing software, sourcing expert quotes, reviewing tools in-depth, and going deeper on topics. The time AI saved in drafting was being absorbed by everything else a competitive post requires.
What made it harder was that some writers felt the pressure of both expectations at once: write faster and write more. When I understood that this was happening, I had to recalibrate. AI hadn’t decreased the workload. In many ways, it had increased the workload.
So my first lesson was: If you’re measuring AI’s impact on your team purely by post count, you’re going to be disappointed. The win is in what those posts become, not how fast they get there.
Research was our biggest surprise.
I also assumed AI would make research faster. After all, instead of spending an hour getting up to speed on a topic, a writer could ask Claude and move on.
For general topic knowledge, that’s actually true. On well-documented subjects with plenty of training data to draw from, AI can explain concepts clearly, answer follow-up questions, and get a writer up to speed faster than navigating a dozen browser tabs.
The shortcomings, however, were flagged by writers almost immediately.
The first time it really surfaced was on a listicle covering competitor products. A writer came back to me flagging that the pricing information AI had surfaced was wrong. Not slightly off, but fully outdated. AI didn’t know the pricing had changed, and it didn’t flag its own uncertainty.
I probably should have seen that one coming. After all, a tool trained on data from months ago obviously can’t know that a competitor changed their pricing. But it took a writer caring enough to flag it before I accounted for it.
From this, we learned that for anything requiring accuracy about competitor pricing or features, AI research falls apart. This makes sense when you think about how these tools work: An LLM’s training data has a cutoff date, meaning it’s unlikely to reflect the latest pricing changes, feature updates, or product rebranding. And when AI tools do attempt a live web search, they’re only as good as the pages they can parse. Product pages are often structured in ways that are difficult for an LLM to interpret, which makes even real-time lookups prone to inaccuracies. The result is information that sounds authoritative but may be months or years out of date.
We produce a lot of competitor listicles and product-focused content, and for those, research still requires significant manual effort. So while AI has sped up general-topic research, competitor- and product-focused research takes roughly the same amount of time as it always has.
The Hallucination Problem
The research problem ran much deeper than product pricing alone. Beyond outdated competitor data, we were seeing fabricated statistics, misattributed quotes, and incorrect terminology, all delivered with the same confident, authoritative tone. AI doesn’t know what it doesn’t know.
We place a high premium on accuracy, and these so-called AI hallucinations turned fact-checking from a lightweight closing step into probably the most time-consuming elements of the method. In some circumstances, AI drafting took longer than handbook drafting as a result of writers needed to spend time verifying what the AI bought mistaken.
What struck me was how shortly my group caught it. That dedication to publishing solely truthful, verified info is precisely what separates a great author from a content material generator. It bolstered one thing I already believed: the author issues greater than the instruments they use.
To handle hallucinations, we launched a brand new formal step in our course of: devoted fact-checking. Pre-AI, writers had been chargeable for the accuracy of their very own analysis, and a closing editorial move would catch something apparent. With AI within the combine, fact-checking turned its personal distinct part. For a typical put up, I estimate it provides round 30 to 60 minutes to the method. It’s longer for research-heavy items or something involving stats, quotes, or product info.
Positive, you may skip fact-checking and assume AI will get every part proper. However is that basically the usual you wish to maintain your content material to? Publishing misinformation, even unintentionally, damages belief in a method that’s exhausting to get better from.
High quality doesn’t police itself.
QA time grew too, for causes past what fact-checking was already catching. Reality-checking is the method of verifying that the data in a put up is right. QA is about every part else: Does this put up sound like an individual wrote it? Does it meet our editorial requirements? Is it structured in a method that serps and AI instruments can parse?
And not using a structured high quality layer, AI-assisted drafts launched issues writers wouldn’t essentially catch on their very own, corresponding to:
- AI clichés and hedge-word-heavy phrasing (“it’s price noting,” “leverage,” “it’s essential to grasp,” and many others)
- Redundant sentences.
- Plagiarism from competing content material (not intentional, however from fashions quoting their supply materials instantly within the output)
- Damaged or defective hyperlinks
- Lacking picture alt textual content
- Incorrect heading orders
- Sentences, paragraphs, and sections that aren’t optimized for LLM chunkability.
To handle this, I constructed a customized QA checker instantly into Google Docs that flags these issues and extra. It’s turn into a required step in our workflow.
What AI Didn’t Contact
By way of all of the modifications above, some issues didn’t transfer. The E-E-A-T necessities we had earlier than AI — unique quotes, firsthand views, strong knowledge sourcing — stayed intact. Editorial judgment stayed intact. The strategic pondering that goes into deciding what a put up must say, and who it must say it to, stayed intact.
AI handles a lot of the execution now: drafting, scoping, producing subject angles, and explaining ideas. However the layer of labor that makes a chunk price studying (the human perspective, the unique perception, the rigorous sourcing) remains to be on the author. If something, it issues extra now as a result of it’s what separates your content material from everybody else’s AI output.
Evaluating Pre-AI vs. Publish-AI Writing
To grasp what modified, it helps to see it. Right here’s a tough breakdown of how time was distributed throughout a typical weblog put up earlier than AI, and the way that distribution shifted after we adopted it:
Observe: Time estimates are averages based mostly on HubSpot Weblog Workforce inside monitoring. Particular person posts fluctuate based mostly on subject complexity, size, and content material sort.
After all, the fact varies greater than these charts counsel. A author would possibly draft for an hour, notice they want extra analysis, return to drafting, catch one thing that wants fact-checking, and return to drafting once more. An actual workflow seems to be extra like this:
Nonetheless, the time financial savings are evident.
AI content material at scale is simple. AI content material price studying is difficult.
If AI can generate a weblog put up in minutes, why not simply publish and transfer on? Extra content material means extra possibilities to rank, extra floor space for leads, and extra output for much less effort, proper?
Early within the AI content material growth, all of us observed a sure technique rising within the search engine marketing area: Publish as a lot AI-generated content material as potential, as quick as potential, and seize market share earlier than rivals caught on. Flood the zone. It operates on the idea that quantity alone could be sufficient to win.
Besides, it isn’t, and the results caught up.
Slop at Scale
The content material that comes from such a fast-and-loose technique is recognizable to anyone who’s spent time in this industry.
A blog post that strings together sentences without saying much. A listicle that could have been written about any company in any industry. No original data. No real point of view. No specific examples that couldn’t have been fabricated. And if you run it through a plagiarism checker, you’ll even find content that was lifted from competing sources (not intentionally, but as a byproduct of how LLMs work).
In short, it’s generic. You’ve probably heard the term “AI slop” to describe this kind of content as well. Experienced readers and, more importantly, search algorithms can spot slop from a mile away, and it hurts far more than it helps.
Marketers are catching on. According to HubSpot’s 2026 State of Marketing Report, 52% of entrepreneurs consider AI has made content material really easy to create that it’s turn into much less efficient total.
Want extra proof? In 2026, SEO expert Lily Ray tracked more than 220 websites that recognized themselves as clients of AI content material creation platforms and located a constant sample: fast development in natural visitors, adopted by a steep decline that, in lots of circumstances, erased the beneficial properties completely. Throughout her dataset, greater than half of the websites misplaced at the very least 30% of their peak natural visitors, and almost 1 / 4 misplaced 75% or extra.
Right here’s an instance of what that appears like for one such web site.
I perceive why groups took to this strategy at first. The strain to supply is actual: 83% of marketers say they’re expected to create more content than ever because of AI.
However that is precisely what main solely with velocity seems to be like in apply. If high quality turns into one thing you attempt to retrofit after the very fact, it hardly ever works.
Begin with requirements, then add AI.
The groups doing AI writing nicely begin from the alternative finish. They outline what “good” seems to be like — voice, depth, accuracy, originality — after which work backward to find out the place AI might help them get there quicker with out compromising on these requirements. All writing is anchored in high quality from the start.
On the HubSpot Weblog, which means maintaining our E-E-A-T necessities intact. We construct high quality guardrails into our workflow and are trustworthy about which content material varieties profit from AI involvement and which don’t. For instance, a how-to information and a chunk of unique analysis can have completely different necessities, and treating them the identical method will end in generic content material that goes unnoticed.
As former Director of the HubSpot Weblog Karla Hesterberg places it: “The one really easy course of produces slop.” Groups that resist the slop, preserve good writers within the driver’s seat, and use AI as a lever reasonably than a alternative are those producing work that converts.
For a extra thorough dive into how we preserve high quality requirements in an AI-assisted editorial course of, take a look at this put up from HubSpot Weblog author Amy Rigby on how HubSpot creates high quality content material within the age of AI.
What This Means for Individuals Managers
Virtually in a single day, my function shifted in two instructions directly. On one facet, I turned extra process-oriented than I’d ever been. I needed to assume extra deeply about workflows, tooling, and the way posts get assembled. On the opposite, folks administration bought tougher. Not as a result of my writers had been tough, however as a result of I used to be asking them to vary how they work.
Course of modifications are comparatively easy to doc, however sustaining motivation and morale by way of a transition that essentially modified their jobs is a special downside completely. You’ll be able to construct probably the most considerate AI writing course of on the earth, but when your writers aren’t purchased in, none of it really works.
Writers come to this job for various causes, and AI lands in another way relying on which motive that’s.
Some writers are wired for outcomes. They get into content material advertising as a result of they’re good at writing they usually like seeing the enterprise affect. For them, AI is a welcome change as a result of it will get them to the end result extra shortly.
Different writers got here to this work as a result of they love the craft. The writing course of is one thing they care about deeply. For them, an AI-first mannequin can really feel prefer it’s taking one thing away. I don’t see this as a efficiency difficulty or a failure to adapt. It’s an inside rigidity that needs to be taken severely.
As a supervisor, my job is to acknowledge the place every group member falls on that spectrum and work with them accordingly. I create an area for trustworthy suggestions, together with damaging suggestions about AI. I do know that if writers don’t really feel comfy telling me what isn’t working, I’ll construct processes on unhealthy info. The writers who most actually pushed again on our early AI workflows gave me the clearest roadmap for what wanted fixing.
Steps for Managing Your Workforce By way of the AI Transition
Right here’s what I’d inform any supervisor navigating AI writing:
- Don’t over-prescribe early. Give writers entry to instruments and room to experiment earlier than locking in inflexible workflows.
- Mandate experimentation, not technique. The groups that prescribed particular AI utilization from day one created pointless friction.
- Acknowledge that you just’re asking writers to vary how they work. That takes time and belief.
- Create area for trustworthy suggestions, together with constructive suggestions. If writers don’t really feel secure telling you what isn’t working, you’ll construct processes on unhealthy info.
- Perceive the place every author falls on the spectrum between outcome-driven and craft-driven, and handle them accordingly.
The aim hasn’t modified: a group that produces work price studying. The trail to get there simply seems to be completely different from what it did in 2024.
My Takeaways as a Content material Workforce Supervisor
A yr into this transition, I’m happy with the place we landed, though we’re nonetheless figuring some issues out.
What made the distinction wasn’t solely the instruments, however the writers. I knew that stepping into, and this previous yr confirmed it at each flip. The writers who pushed again actually gave me the roadmap, and those who struggled helped me perceive what was really being requested of them.
What sticks with me most from this expertise is one thing writers have began saying just lately: that they’re feeling like actual writers once more. The steadiness between AI effectivity and precise craft has began to really feel sustainable. That occurred not as a result of we discovered an ideal immediate or the appropriate instrument, however as a result of we by no means let go of what made our content material good within the first place.
In the event you’re an editorial lead navigating the AI shift proper now, know this: What will get you thru is having writers who care sufficient to be trustworthy with you, and being a supervisor who creates area for that honesty. Don’t underestimate what it means for a author to really feel like a author. It exhibits up within the work.
AI is a instrument. Maintain your writers on the heart, maintain the standard line, and the remainder follows.
