OBJECTIVES: Large language models (LLMs) are widely used for research tasks; however, they require human expert input to create suitable prompts for accurate outputs. This raises questions about their cost-effectiveness. We have previously reported the accuracy of LLMs in indexing abstracts by disease area and hallmarks of ageing. Now, we aim to evaluate the costs of both manual and LLM-assisted indexing approaches to determine the most cost-effective method.

METHODS: Using an online evidence mapper tool (www.evidencemapper.co.uk), we classified 500 abstracts on anti-ageing treatments by disease area and hallmarks of ageing. We calculated the cost of manually indexing and verifying abstracts by considering the opportunity cost, using indicative hourly rates of £130 for researchers and £150 for senior researchers. For LLM use, costs included software fees, the opportunity cost of the expert generating the prompt (at an indicative cost of £280 per hour), and the time researchers spent verifying the LLM's indexing. Costs were stratified by task complexity: disease indexing vs. hallmarks of ageing indexing.

RESULTS: Manually indexing abstracts by disease area cost £2,197 for 16.9 hours of work. Using an LLM cost £2,171.35, including £840 for the expert, £1,312.50 for researchers, and £18.85 for software. For the more complex task of indexing by hallmarks of ageing, manual indexing cost £7,598 (55.8 hours). Using an LLM cost £3,348.85, comprising £1,120 for the expert, £2,210 for researchers' verification, and £18.85 for software.

CONCLUSIONS: Our study demonstrates that there can be an economic advantage of using LLMs for repeated indexing tasks. The cost-savings are greater where the data to be indexed is more complex as human researchers need longer to process the data. The time taken to set up an accurate prompt suggests that LLMs may be more cost-effective for larger data sets, as the marginal cost of indexing additional abstracts is small.