If you are sceptical about the impact of policy documents, whether read or filed in the wastepaper basket, you should consider the case of assertive outreach.

It is mentioned as one of several approaches to organising effective community treatment in paragraph 4.21 of Modernising Mental Health Services . Better outreach is also one of the six types of service listed in the table summarising the policy under the heading of 'safety'. That's it.

Since then, assertive outreach teams have appeared everywhere, especially in inner cities. Job sections of journals are dominated by advertisements for assertive outreach workers, team leaders and managers.

The new popularity is easy to understand. After all, the model is evidence-based, and deals with many of our concerns. Assertive outreach promises to engage the people no-one could reach before - especially psychotic, drug-abusing, alienated, dangerous men.

These teams make a major contribution to safe care and reduce length of admission, paying for themselves in the process. In other words, they promise social inclusion and safe communities at lower cost.

In our dreams, all this is achieved by perhaps eight staff - a single team which is a very small, and mostly marginal, part of a local mental health service.

It is intriguing how the expectations of mental healthcare as a whole have come to be projected on to these teams. It is not a case of the emperor without his clothes - after all, the system does work in the right circumstances - but the emperor wearing his whole wardrobe. The emperor will be overburdened and poorly colour-co-ordinated, and may predictably collapse under the weight.

Many hopes and concepts have been joined together in the new assertive outreach policy. It may appear irresistible as an ideal, but its tenets could prove irreconcilable in practice.

Two models dominate. In caricature, the model somewhat unfairly attributed to the government portrays assertive outreach as controlling patients who are posing a risk. Staff will identify them, gain access, impose treatment and force compliance (if that is not a contradiction in terms). If medication and other interventions are not achieved by persuasion, coercion will be supported by the proposed compulsory community orders, and hospital will be used as a place of restraint where medication can be administered involuntarily.

It is a professional model put into practice by staff with service-driven objectives. Risk - and especially homicides - have to be eliminated, and the priority is the safety of local communities. Such services will not be popular with patients, who will resist involvement, and will have to be hunted down in communities. Too stark? Remember the wall of police in riot gear on television trying to flush out a single vulnerable person?

The alternative model is the one proposed by organisations such as the Sainsbury Centre and the King's Fund. This is squarely aimed at the service user's agenda, attempting to achieve engagement by starting from the perspective of people's lifestyles and ambitions, and considering their interaction with family and communities.

Often, staff will have been recruited from the same social and cultural background as the targeted service users. Assertive outreach will link people to essential support services such as benefits, housing, occupation and treatment on their terms - but there are boundaries to tolerance.

Therefore, risk assessment is important as an aspect of overall needs assessment.

This model assumes that many users are alienated from traditional statutory services that have little to offer them beyond symptom control.

Instead, assertive outreach should be the stylus in the record player - a small but essential part of the system.

But unless contact is optimal and fits the groove perfectly, the music may turn ugly, and however upmarket the rest of the system, it is useless. Too liberal and wishy-washy for these cynical times? Maybe, but we also live in an evidence-driven age, and it can work.

So we have a risk-minimisation and a social integration model - mutually exclusive, but we want them both.

Intriguingly, these models, though not at ease with each other, reflect tensions in society at large. Think of our approaches to single mothers, and to homeless and unemployed people. All have to cope with a simultaneous barrage of incentives and sanctions which are not always well integrated.

The choice between the two ideologies does matter, since it conditions public expectations, staff behaviour and user responses.

Imagine you are a reluctant patient, who speaks English poorly, with nothing to do all day but smoke the odd cigarette, and not keen to take medication. It makes a difference if a knock on the door means the riot squad, showing zero tolerance, or a person sharing your values, exploring ways to help.

It is equally important that staff know whether they will be ultimately judged on delivering a risk-free zone as defined by inquiries, or satisfied customers as measured by a national survey.

Society will have an opinion, but government has to decide on its behalf. The worst option is to leave it all hanging in the balance, hoping for locally brokered compromises that miraculously deliver all.

This will not happen, not just because local planners will refuse to take the poisonous bait, but also because the approaches require different attitudes, knowledge and skills from staff, influencing training programmes, and attracting very different recruits.

There is a very small window of opportunity in which we can make these choices. If we want to have effective care of either variety, choices have to be made now, and communicated to those implementing and delivering mental healthcare.