The news comes in the REF panels' draft panel criteria and working methods, put out for consultation last week by the Higher Education Funding Council for England, which manages the exercise on behalf of all four UK funding councils.
Of the 36 REF subpanels, 12 intend to use citation counts, including all six bioscience subpanels and four from the physical sciences: chemistry, physics, earth systems and environmental sciences, and computer science and informatics.
Citation information will also be used by two social science subpanels: geography, environmental studies and archaeology, and economics and econometrics.
But all panels have emphasised that citation data will be used only to inform, rather than replace, the peer review of outputs. The data will be drawn from standard citation databases, although the computer science subpanel will also be permitted to use Google Scholar.
The subpanel's chair, Steve Furber, ICL professor of computer engineering at the University of Manchester, said experience of the 2008 research assessment exercise suggested that using both databases was the best way to take account of some of the differences in publication behaviour in computer science, such as the importance of conference papers, which are less well indexed by standard databases.
"We were also of the collective view that the community we are assessing will have more confidence in the process if we look at (both)," he added.
Each main panel's draft criteria also set out the kinds of evidence it will accept for the impact element of the REF, which will count for 20 per cent of final scores.
The research underpinning impacts must have occurred since 1 January 1993 except for the architecture, built environment and planning subpanel, whose cut-off will be five years earlier to reflect the length of time it can take for "changes to the built environment to be delivered in practice".
Academics are also invited to comment on a number of general issues, including the circumstances in which researchers may submit co-authored entries or fewer than the standard four outputs. Responses must be received by 5 October.