37signals state that they don't use personas because:

"We use ourselves. I believe personas lead to a false sense of understanding at the deepest, most critical levels."

Personas are poor substitutes for real people, they argue, and you should always base your decisions on the problems and behaviour of real people instead of passive abstractions of people. This reasoning fits perfectly with their "getting real" process, which argues for a very agile development process.

It's an interesting point, but this doesn't mean that personas are therefore useless. Of course, personas should never be used as stand-ins for actual users, and thus can never provide the feedback that real people provide - it's impossible to do usability testing with them!. Rather, they're a heuristic tool to contextualise your design problems. Personas allow you to frame your design in a way that puts the goals of real people at the epicenter of your interface. By making personas an integral part of the story that your design tells, they become a common ground for all involved parties, stakeholders, developers, designers, managers.

Personas should be based on proper, qualitative user research. The data can come from various sources, eg. interviews, discourse analysis, observations, surveys etc. The analysis is actually the hardest part, since it's very difficult to step outside your own frame of mind when trying to really understand people's behaviours. The ethnographic techniques can be learned quite easily, but the proper understanding of the methodology and the required reflective use of it makes it a very difficult task without academic education.

Clay Spinuzzi's observation that designers often frame themselves as 'heroes' to rescue the poor 'users' is a perfect example of designers executing a naive, superficial user research. But the solution is not to use yourself as the intended user of your application, unless you're building a product which is only meant to be used by yourself. Doing the proper research and crafting personas and scenarios are a good first step in designing not just for yourself. However, even 37signals can't be as arrogant as to presume they're representative for all the possible people that their products intend to reach. As with any tool or methodology, using it without proper understanding ultimately leads to failure, but that shouldn't be a reason to dismiss the tool completely.

Edit 18/11/07: Other people weigh in as well.

• Jared Spool makes a similair point on the difference between crappy personas vs. robust personas


• Boxes and Arrows publishes a timely article on building a data-backed persona


• Andy Budd chips in with a post on rethoric sensationalism