For the past two years, the dominant advice in B2B content marketing has been some version of: be careful with AI. Don’t let it water down your brand. Keep a strong human in the loop. Avoid the slop.
That advice made sense when it was speculative. It makes less sense now that we have data.
Semrush recently analyzed 42,000 blog posts, human-written, AI-generated, and mixed, and tracked where each type lands in the SERP.
Human content outperformed at position #1: 80.5% of top-ranking pages were human-written vs. 10% AI-generated.
But from position 5 onward, the gap narrows sharply, which means if your goal is page-one visibility, AI content competes.
So if AI content isn’t the problem, what is?
Table of contents
The real risk isn’t quality. It’s voice drift.
Scaling content has always been a consistency problem. Hire more writers, and you get more variation. Hire cheaper writers, and you get lower quality. Add AI, and suddenly you’re publishing at 3x the volume with 3x the chances of sounding like everybody else.
Voice drift is what happens when no one’s monitoring brand content as output increases. One post sounds sharp and opinionated, while the next sounds like a LinkedIn thought leader on autopilot.
The fix: a tighter feedback loop between how you write and what your LLM outputs.
The problem with most AI content isn’t that it’s AI, it’s that it’s trained on the average of the internet. If your brand voice is anything other than average, you have to teach an LLM your tone of voice explicitly.
Feeding a generic prompt into Claude or ChatGPT, to get decent-ish output, then spending 40 minutes humanizing AI content or editing it back into something that sounds like you, isn’t a system. That’s manual correction with extra steps.
Analyze your own content before you prompt anything.
Before you can teach an LLM your tone of voice, you need to know what your voice actually is: not the version in your brand guidelines, but the version that shows up in your best-performing posts.
Those are often very different.
While brand guidelines describe intentions, published content reveals patterns:
- What sentence structures do you use?
- How long are your paragraphs in practice?
- Do you lean on analogies, statistics, or direct commands?
- Do you ask questions or make declarations?
A tool that scrapes your recent posts and surfaces these patterns will tell you more about your real voice than any brief your marketing team has written. That’s the baseline. Work from that, rather than an idealized document.
If you have a tone of voice guide, don’t discard it—compare it.
Where your documented voice and your published voice diverge, you have a problem worth solving before you scale anything.
Turn your voice analysis into a prompt your LLM can use.
The next step is to translate your voice into something that changes AI output.
“Write in a conversational but professional tone” does nothing. Every LLM defaults to something it calls “conversational but professional.” What you need is specific, behavioral instruction—not adjectives, but patterns.
Good LLM tone of voice prompts include:
- Sentence structure rules: “Lead with the insight, then explain. Never bury the main point in paragraph three.”
- Formatting constraints: “Short paragraphs. Two sentences, three at most. Use single-sentence paragraphs for emphasis.”
- Vocabulary bans: “Don’t use ‘leverage,’ ‘synergy,’ ‘unlock potential,’ or ’empower.’ These are off the list.”
- Proof requirements: “Every claim needs a specific example. No generic assertions.”
- Tone anchors: “Write like you’re explaining something to a smart colleague who’s skeptical, not a prospect you’re trying to impress.”
The more behavioral your prompt, the more your output reflects your actual voice.
Mastering your tone of voice and teaching it to your LLM is key, especially if you aim to scale content production consistently.
So we built a blog voice analyzer tool that does exactly that.
To build your own voice analyzer, you can also find the prompt used to build this app here.
If you’re using Claude Code, you can go further: export your analyzed posts as .md files and include them as direct context.
This gives your model real examples and much richer context to work with.
And if you already have your tone of voice documented, you can paste it into the tool.
It will compare it with the generated report, show how closely they match, highlight any gaps, and suggest how to refine it even further.
What to do next
Getting your LLM to write in your tone of voice isn’t a one-afternoon project, but you can move fast.
- Audit your five best posts: Pick five pieces you’d show a new hire as examples of your brand at its best. Look for patterns: average paragraph length, how your H2s work—questions or declarations—and your ratio of data to opinion. This is your voice fingerprint.
- Run a scrape-and-analyze pass: Use a tool that processes your recent content and identifies structural patterns automatically. The goal isn’t a perfect brief; it’s a starting hypothesis you can refine. Compare what you find against whatever brand guidance you already have.
- Build a behavioral prompt, not an adjective list: Take the patterns you found and turn them into constraints. Be specific. “Two-sentence paragraphs” beats “punchy.” “Lead with the counter-intuitive finding” beats “engaging.” Test this prompt on a piece you’ve already written and see how close the output gets.
- Build a human review step: AI drafts, you humanize the AI content—but edit for voice, not just accuracy. Every time you change something, ask why. Those corrections are data. Feed them back into your prompt.
Make your LLM tone of voice a system, not a style guide.
The scaling content problem has been solved. Tools can write faster than any team can publish. That’s not the constraint anymore.
The constraint is whether any of that output actually sounds like you: is it specific enough to be recognizable, sharp enough to be worth reading, and consistent enough to build trust over time.
Generic AI content loses that battle before it starts. An AI workflow using LLMs trained on your patterns, your proof points, and your actual published work is the differentiator. It’s a production system that compounds rather than dilutes.
If your content still depends on manual effort, inconsistent outputs, or endless editing to “make it sound right,” CXL’s B2B Marketing and AI and AI Agents for B2B Marketing programs are built to help you learn how to operationalize AI properly.
For live and on-demand AI-focused sessions, check out these courses and more:
- Measuring the modern AI-powered funnel
- Build content automations for lead generation with n8n
- SEO automation with n8n and Claude Code
- B2B buyer pain-led GTM strategy
- B2B ad campaigns with Claude Code and n8n
- Optimizing B2B content funnels with AI
- Build apps as content with AI
- AI-ready agencies: what clients see
- Unique positioning for agencies
