Most conversations about AI focus on what AI knows. This one is about what you know - and why it matters more than most people expect.
There's a particular kind of confidence that AI projects that catches people off guard. Ask Claude, ChatGPT, or Gemini a question and you'll get an answer delivered without hesitation, without caveats, and without the slight pause a person might use to signal they're not entirely sure. It sounds authoritative. And it often is.
But sometimes it's completely wrong.
Chanel Turner-Ross, Head of Marketing at Utopi, flagged this on the first episode of The AI Method Podcast. Her team operates in environmental and building data - a space where numbers have to be accurate. There are times, she said, where calculations or statistics have come back from AI that were "absolutely made up" - incorrect content that could have been published on external channels if no one had caught it.
But Chanel and her team did catch it, because they were experts in their domain.
The question worth asking is: would your team do the same?
| Key Facts: AI Hallucinations and Domain Expertise | |
|---|---|
| Hallucination rate | Studies suggest large language models produce factual errors in 3-27% of responses, depending on task type and domain |
| Confidence mismatch | AI systems do not reliably signal lower confidence when producing incorrect answers |
| Detection advantage | Users with domain expertise catch AI errors significantly more often than those without it |
| Highest-risk PBSA content | Sector statistics, energy performance data, occupancy benchmarks, compliance information |
What is an AI hallucination, and why does it matter for PBSA marketing?
An AI hallucination is when a large language model produces information that is factually incorrect but presented with the same confidence and fluency as accurate content. For PBSA marketing teams producing content about occupancy rates, energy performance, or sector benchmarks, a single unchecked hallucination can result in incorrect claims reaching residents, clients, or regulators.
AI hallucinations occur because language models generate responses by predicting statistically likely sequences of words - not by retrieving verified facts from a database. The model has no mechanism for uncertainty that maps onto what humans would call doubt. A made-up statistic delivered in a clean, well-formatted paragraph looks identical to an accurate one. (I teach this in my Tool to Teammate training session).
For general or subjective content - a welcome message, a caption, a community newsletter - this is a manageable risk. For sector-specific numerical claims, it isn't. The PBSA sector operates in a space where data carries weight: with investors, with regulators, with residents, and with the operators who are increasingly benchmarking performance across portfolios. Inaccurate figures in that context aren't just embarrassing - they can erode trust or, in compliance-adjacent content, create real exposure.
Why does sector expertise act as your quality control layer?
PBSA marketing professionals who understand their sector can identify when AI-generated content contains inaccurate statistics, implausible benchmarks, or claims that contradict known market conditions. This expertise functions as a quality control layer that no AI tool can replicate, because the AI has no access to your portfolio's actual performance, your internal data, or your direct experience of the sector.
"If it tells you something, it's so confident that you just assume that it's correct. Because if you speak to a person, the person will maybe hesitate... Whereas an AI would just be like, yes, here's the answer...even if it's wrong." - Ross Gledhill, Director of Engineering, Utopi
The point Ross is making is that human scepticism - the natural response of someone with experience - is the check that AI itself cannot provide. A marketing manager who has spent years working with occupancy data will notice when a figure looks implausible. Someone newer to the sector, or working in an area where they're less confident, may not. That gap is where errors get published.
Chanel put it simply on the podcast: no one knows your job better than you do. That's not a modest observation - it's the specific thing that makes an experienced PBSA marketer a more effective AI user than someone technically proficient but sector-naive.
Which types of PBSA content carry the highest risk?
Content types that carry the highest risk of undetected AI errors in PBSA marketing are those involving specific numerical claims: occupancy and void rates, energy performance figures, student satisfaction statistics, and regulatory compliance information. These are areas where AI is more likely to produce plausible-sounding but incorrect data - and where incorrect data carries real reputational or legal risk.
Broadly, the PBSA content areas most vulnerable to hallucination risk include:
- Occupancy and void rate statistics, particularly sector-wide benchmarks or city-level comparisons
- Energy performance data and net zero claims, which are subject to regulatory scrutiny and investor reporting requirements
- Student satisfaction figures and NPS comparisons across operators
- References to compliance requirements such as fire safety regulations, legislation, or planning conditions
- City or region-specific market data including average rental levels and supply pipeline figures
The inverse is equally useful to know. Content involving subjective judgements - tone of voice, creative framing, community storytelling - carries lower risk because there's no verifiable correct answer for the AI to get wrong. Directing AI effort toward creative tasks and keeping factual claims human-verified is a practical way to manage the risk without slowing down production.
How does expert-led AI use look different in practice?
Marketing teams who use AI most effectively treat it as a capable but unsupervised junior colleague - one who produces work that needs to be checked before it goes anywhere. They verify any specific data point AI generates, particularly statistics and compliance references, against a primary source before including it in external content.
They also brief with scepticism built in. Prompts like "note where you're uncertain" or "do not include statistics you cannot source" don't eliminate hallucinations, but they reduce the likelihood of AI confidently producing invented figures in areas where accuracy matters.
They know where their own knowledge is thinner. If your team is producing content about energy performance legislation but has limited direct familiarity with the underlying regulations, that's a gap worth closing before relying on AI to handle that content area. The value of AI in those cases is in drafting structure and language - not in supplying facts.
What does this mean for PBSA teams thinking about AI adoption?
PBSA marketing teams should treat domain expertise as a competitive advantage in AI adoption, not as something AI will make redundant. The professionals who know their sector best are best placed to use AI most effectively - because they can catch errors, ask better questions, and direct AI output toward results that are accurate as well as efficient.
The framing that AI is coming for marketing jobs misses something important. The marketing professionals most at risk are not the most knowledgeable ones. They're the ones who use AI without applying the judgement their experience gives them.
The teams making the most progress with AI in student accommodation aren't the ones who trust it most. They're the ones who know their subject well enough to push back when something doesn't look right. That combination - sector expertise plus the right tools - is where the results are.
About The AI Method: Oliver Harrison works with in-house Content and Marketing teams, helping them move from 'thinking about AI' to actually working with it. If you'd like to explore how AI tools could fit your team's workflow, find out more at www.theaimethod.co.