If your content team is publishing more AI-assisted articles and your AI visibility numbers are rising, it is tempting to conclude the strategy works. A growing body of research suggests a less comfortable reading: answer engines may be rewarding machine-written content partly because it is machine-written, and the metrics most teams watch will not show the damage. For B2B brands, the durable response is content that a model cannot produce on its own.
The argument
In Search Engine Journal, Duane Forrester, founder of UnboundAnswers, argues that AI answer engines show a measurable preference for AI-generated content. That creates a loop: synthetic content gets cited, so more synthetic content gets produced, so the pool the engines draw from becomes more synthetic. Citation counts and traffic can look healthy throughout, which is why he says the problem sits on a dial almost nobody watches.
Related: Unlocking B2B Content Marketing Strategies for Leadership Teams (2026)
The research he draws on
Forrester brings together several strands of work. In our own summary:
- An analysis by Graphite found that more than half of newly published English-language web articles are already AI-generated.
- A study presented at SIGIR described an "invisible relevance bias", where retrieval systems ranked AI-generated items above human-written ones without a relevance reason for doing so.
- A 2026 Web Conference paper modelled "retrieval collapse": once synthetic content made up about two-thirds of the pool, more than 80% of retrieved sources were machine-written.
- Research published in Nature showed that models trained repeatedly on generated data degrade over successive generations.
Each finding has its own limits, and none of them measures your specific market. Together they describe a plausible mechanism worth taking seriously.
Why the dashboards look fine
The uncomfortable part is that none of this shows up in standard reporting. Your pages are indexed, cited and visited. What you cannot see is whether the answer pool in your category is narrowing, whether every vendor is being described with the same recycled phrases, or whether buyers trust those answers less over time.
For a B2B company, that matters in two ways. First, if your content looks like everyone else's, you are interchangeable in an AI answer. Second, if the pool your category draws from fills up with generic material, the vendor that supplies genuinely new information becomes the most valuable source in it.
Related: Mastering Content Strategy: Tools to Automate Your Way to Success (2026)
What B2B teams should do differently
Forrester's recommendations translate well into a B2B content plan. Our practical version:
- Publish original evidence. Anonymised data from your own product, survey results, benchmarks you ran yourself, and documented customer outcomes with permission. A model can paraphrase existing articles; it cannot invent your data.
- Make authorship verifiable. Real names, roles and relevant experience on every substantial piece, with links to the author's profile.
- Show your sources. Link to the studies and pages you rely on, as this article does. Checkable claims are more trustworthy to readers and easier to cite.
- Watch source diversity. When you track AI answers for your key prompts, record which sources are cited, not only whether you are. A shrinking set of sources is an early warning.
- Avoid the generic template. If your articles share the same structure, openers and phrasing as every other AI-assisted post, they add nothing a model does not already have.
Where AI still belongs in the process
None of this means abandoning AI writing tools. They are useful for research, outlines, editing and repurposing. The problem is using them to produce the substance. A practical rule for content teams:
- AI can help shape and polish the piece.
- The core facts, examples and conclusions must come from a person with direct knowledge or from verified sources.
- Every number must trace back to a source you can link or a dataset you own.
A quick audit of your own content
Before changing strategy, check where you stand. Take your ten most important articles, the ones that target buying questions, and ask four questions of each one. Does it contain at least one fact, example or number that exists nowhere else? Is a named author with relevant experience attached? Are the claims linked to sources a reader can check? Could a competitor publish the same piece by swapping in their own brand name?
Articles that fail most of these are the ones most exposed to the loop Forrester describes. Rewrite them first, starting with the pages closest to a buying decision, such as comparisons, pricing explanations and implementation guides.
First-party data is the moat
The most defensible B2B content is built on data only you have. That can be product usage patterns, results from your own campaigns, or aggregated trends across customers, shared carefully and without identifying anyone.
It also applies to how you understand your audience. Knowing which companies actually read your content, rather than inferring it from traffic totals, tells you which topics attract buyers. Website visitor identification with Happierleads shows the companies and, where privacy law allows, the people behind visits to each article, so your editorial choices rest on real demand instead of volume metrics.
The takeaway
The web filling up with machine-written content is not a reason to panic, but it is a reason to stop measuring content success only by output and citations. Invest in material a model cannot generate: original data, named expertise and checkable sources. Track the diversity of what gets cited in your category, not just your own share. When the pool gets more generic, the brands that bring new evidence become the ones worth citing.
