The Thomson Reuters 2025 Generative AI in Professional Services Report found that 26% of legal organisations are now actively using generative AI, almost double the 14% the same survey reported a year earlier. More striking: 78% of legal professionals expect generative AI to be central to their workflow within five years. That is the biggest gap between current usage and predicted future usage in any professional services category Thomson Reuters tracks.
For legal assistants and paralegals, that gap is the entire story of the next two years. Document review, legal research, and document summarisation, which the same report flags as the top three use cases (cited by 77%, 74%, and 74% of respondents respectively), make up most of what a paralegal’s week looks like. Those are the workflows that change first, change most, and change visibly to the attorneys above them.
The Anthropic Economic Index puts paralegals in the top tier of exposure
The Anthropic Economic Index consistently flags paralegals and legal researchers as among the highest-exposure occupations to generative AI. That label gets read two ways depending on who is reading it. Pessimistically, it is a sign the work is going to be automated away. Optimistically, it is a sign the work is exactly the work AI is well-suited to assist on.
The actual usage data favours the second reading. The same Index shows augmentation patterns (52% of conversations) outweigh automation patterns (45%) across roles, including the high-exposure legal ones. Translated: paralegals and legal assistants who are using AI are using it as a research-and-drafting partner, not as a replacement for their judgement. The work that goes through AI on the way to being checked, edited, and signed off by the human is more work than the work that goes through AI on the way to being shipped without human review.
The economic picture, sourced
The Thomson Reuters report estimates 5 hours per week saved within a year of meaningful AI adoption, equivalent to about $19,000 per professional annually at typical legal billing rates. The US Bureau of Labor Statistics projects lawyers to grow 4% from 2024 to 2034, with about 31,500 openings every year and around 864,800 lawyer jobs in 2024. Paralegal and legal assistant data follows a similar pattern: continued growth, no projected decline, with the role increasingly involving AI-assisted work rather than being displaced by it.
Importantly, no large law firm has publicly announced paralegal headcount reductions on the basis of AI deployment. The pattern instead is firms hiring the same number of paralegals and asking them to operate at a higher rung of the work, with first-pass document review and summarisation done by AI under their supervision rather than by them directly.
What changes inside the role
Three shifts are visible across firms that have moved past pilot stage on generative AI.
Document review changes from being the work to being the supervision of the work. A paralegal sitting on a litigation matter no longer reads every document for relevance. They define the relevance criteria, run them through a tool like Harvey, Casetext CoCounsel, or general-purpose models with appropriate confidentiality controls, then audit the output for the boundary cases the model got wrong. The cognitive load shifts from volume to judgement.
Legal research changes from “find me cases on” to “find me cases on, and tell me what they say, and where the contradictions are.” Westlaw Edge, LexisNexis, and the newer AI-native tools have moved research from a search-and-list interface to a question-and-summary interface. Paralegals doing the research work the older way are visibly slower than the ones doing it the new way.
Drafting changes from “write the first version” to “edit the model’s first version.” First drafts of standard documents (engagement letters, basic motions, discovery requests, deposition summaries) are now routinely model-generated and then edited rather than written from scratch. The skill that matters is knowing what to keep, what to cut, and what to flag for an attorney.
Where the human work concentrates
The work that is not going through AI is the work that requires institutional knowledge of the matter, judgement on what the partner actually wants, knowledge of the client’s preferences and history, and the trust required to be told things the client does not want in writing. None of that is in the AI’s training data, none of it is going to be, and the parts of the role that depend on it are the parts where the senior paralegal earns their seniority.
The other category of work the AI does not touch is the procedural and physical layer of legal practice: filings, court appearances, witness liaison, exhibit management, the small but important rituals that hold a matter together. That is unglamorous work but it is reliably present, and it has not been automated.
What the next 12 months look like
The Thomson Reuters projection of 78% of professionals expecting AI to be central within five years is the directional anchor. Adoption inside firms will keep climbing through 2026, with mid-sized firms catching up with the large-firm leaders, and confidentiality-and-data-governance frameworks settling into more recognisable patterns.
Three things to watch over 2026. First, billing structure tension. Most firms still bill by the hour, and AI compresses hours. The firms that move fastest to alternative fee arrangements (fixed price, value-based, success-based) will turn the productivity gain into margin. The firms that resist will hand the gain to their clients. Second, the bar regulators in major jurisdictions will publish more concrete guidance on supervision, confidentiality, and citation-verification requirements (driven in part by the early embarrassments of attorneys submitting AI-fabricated case citations to courts). Third, the gap between paralegals who run AI workflows and paralegals who do not will widen visibly, with the AI-fluent group moving up into more substantive work.
If you work in legal support, the question is not whether the role survives. It does. The question is whether you are the paralegal whose attorneys cannot run their week without you, or the one whose remit is steadily compressed back into the work AI is not yet good at.