THE PROPERTY FILTER TAKE
On 18 September 2026, Property Vision partner Rob Fanshawe warned that buyers relying on AI property research could make errors costing thousands of pounds.
The numbers are pointing to a data problem: an AI answer is only as good as the comparables it reads, and asking prices are not sold prices.
Consider checking every AI-generated comparable against HM Land Registry sold prices and the EPC register before you make an offer.
Buyers who rely on AI property research risk errors running into thousands of pounds, according to buying agent Rob Fanshawe, partner at Property Vision (PropertyWire, 18 September 2026). His point is not that AI is useless. It is that a confident answer built on the wrong comparables (recent sales of similar nearby homes, used to judge a price) can look like analysis when it is not.
What did the buying agent warn about?
Fanshawe set out a hypothetical example, reported by The Negotiator on 18 September 2026. A buyer considering a £6 million house in the Cotswolds asks an AI platform whether the asking price is justified. The answer says the house looks expensive against three nearby homes, all apparently supporting a lower price per square foot.
Then the comparables fall apart. One is an asking price rather than a completed sale. Another required a full refurbishment, and the third sits on a busy road. "Meanwhile, the genuinely relevant transaction - a similar house sold privately two streets away - never appeared on a portal at all," Fanshawe said (The Negotiator, 18 September 2026).
So the data in his example fails on every count. Each comparable fails for a different reason, and the one sale that mattered was missing entirely. Fanshawe's own verdict: "The analysis looks impressive. The conclusion is misleading."
Where does AI property research go wrong?
According to Fanshawe, much of the problem is that the information AI interprets was imperfect to begin with. "AI can confuse asking prices with achieved prices, recycle old commentary as though it describes today's market, or draw sweeping conclusions from postcode-level data that bear little relation to a particular street or house," he said (The Negotiator, 18 September 2026).
The first of those matters most for pricing. An asking price is what a seller hopes to get, compared to a completed sale, which is what a buyer actually paid. An analysis that mixes the two is measuring different things. For more on assessing a deal, see our guides to analysing property deals.
Fanshawe also flagged the speed of the answer itself. "Today they can receive a beautifully structured answer in under 10 seconds and assume the uncertainty has disappeared. It hasn't," he said.
But this is not a case against AI. "None of this is an argument against buyers being well informed, or against AI, we use it ourselves," Fanshawe said. He added that AI can interrogate large quantities of information, summarise documents, research planning histories and flag questions worth investigating further.
How can investors check AI research against primary data?
The underlying picture is that AI output is a starting point, and primary records are the check. HM Land Registry publishes Price Paid Data, which contains the sale prices of properties in England and Wales submitted to it for registration. That is the record of what homes actually sold for, not what they were marketed at.
For a single property, HM Land Registry's search service gives a free property summary, and a title register costs £7 (GOV.UK). The title register usually includes how much the property was last sold for and who owns it, covering England and Wales only.
Energy efficiency claims can be checked too. The government's find an energy certificate service lets you search for an EPC (energy performance certificate, which rates a property's energy efficiency). You can search by postcode, street name and town, or certificate number. It covers England, Wales and Northern Ireland, and Scotland has a separate service.
It may also be worth running your own numbers rather than accepting a headline figure. Our free property investment calculators let you test a price against your own assumptions.
Fanshawe's closing line frames the test well: the advantage "will belong to whoever can distinguish intelligence from noise" (The Negotiator, 18 September 2026). For the wider picture, see our property investment strategy guides.
Key takeaways
On 18 September 2026, Property Vision partner Rob Fanshawe warned that relying on AI property research could cost buyers thousands of pounds.
In his hypothetical £6 million Cotswolds example, all three AI-selected comparables were flawed and the most relevant private sale was missing.
AI can confuse asking prices with achieved prices and stretch postcode-level data to a single street, according to Fanshawe.
HM Land Registry Price Paid Data contains the sale prices of properties in England and Wales submitted for registration, and a title register costs £7 on GOV.UK.
Fanshawe does not dismiss AI: his firm uses it, and he says it can summarise documents and research planning histories.
Frequently asked questions
Can I trust an AI valuation of a property?
Where can I check what a property actually sold for?
How do I check a property's EPC?
