Four out of five children can't tell when an adult is posing as a child online, according to software developers working to track pedophiles online.
The team, at Lancaster University, has created a tool which can work out a person’s age and gender using language analysis techniques.
The researchers hope it will eventually be used to help law enforcement agencies spot when an adult is 'grooming' a child.
over several months, 350 schoolchildren unknowingly took part in the experiment, in which they chatted to others online - with some of the other 'children' being replaced by adults aiming to deceive.
Even pupils as old as 17 struggled to tell the difference, and overall only 18 percennt of children guessed correctly.
Girls were better than the boys, with 22 percent guessing correctly, compared with 16 percent of the boys.
The team's language analysis software did significantly better correctly working out whether web chat was written by a child or an adult in 47 out of 50 cases.
"In our analysis we found that four out of five children across the school got it wrong," said lead researcher professor Awais Rashid said.
"Interestingly the strategies they use to detect who they are talking to are also the ones that lead them to make wrong decisions – they rely on the subject matter, use of slang and even something as simple as whether the individual said he or she was an adult or a child."
Eventually, the researchers believe the software could be used not only to identify adults posing as children but also to pick up on the 'stylistic footprints' of paedophiles, and track them on the internet.
"We are concerned about the dangers our pupils face when they’re online and have welcomed the opportunity to help the project at the same time as raising pupils’ awareness of the risks," said the head of the school involved in the experiment.
"It has been chilling to watch them being taken in by adults masquerading as teenagers."