Editor’s Note: This article is the sixth post in our multi-part series on the Summer 2025 eDiscovery Pricing Survey, conducted by ComplexDiscovery OÜ in partnership with the EDRM (Electronic Discovery Reference Model). The semi-annual survey drew input from 70 legal, business, and technology professionals, offering a snapshot of how generative AI is being adopted — and priced — in eDiscovery.
Here, we turn to GenAI-assisted review, an area that is both highly promising and highly unsettled. The results underscore both experimentation and uncertainty, reflecting a market that is still searching for reliable benchmarks.
Content Assessment: Generative AI at the Frontier of eDiscovery - Pricing Insights from the Summer 2025 Survey
Information - 93%
Insight - 90%
Relevance - 93%
Objectivity - 94%
Authority - 95%
93%
Excellent
A short percentage-based assessment of the qualitative benefit expressed as a percentage of positive reception of the recent article from ComplexDiscovery OÜ titled, "Generative AI at the Frontier of eDiscovery: Pricing Insights from the Summer 2025 Survey."
Industry Research
Generative AI at the Frontier of eDiscovery: Pricing Insights from the Summer 2025 Survey
ComplexDiscovery Staff
If forensics is the front door, processing the engine room, and review the human core of eDiscovery, then generative AI represents the uncharted frontier. It is where technology begins to move beyond assisting human reviewers to actually transforming how discovery is conceived and conducted.
The Summer 2025 survey highlights just how unsettled this frontier remains. Pricing models for GenAI-assisted review lack consensus, with providers and clients experimenting across per-document, hybrid, and outcome-based approaches. The uncertainty is striking: nearly half of the respondents expressed that they did not know or found questions inapplicable when asked about GenAI pricing.
Just as review remains the human core of eDiscovery — accounting for the largest share of costs and workloads — GenAI introduces a new dimension that could entirely reshape it. The parallels are clear: where review pricing today stabilizes around hourly and per-document norms, GenAI pricing is still in flux, with the potential to rewrite both models and expectations in the years ahead.
Primary Models for GenAI-Assisted Review
When asked about their primary pricing model for GenAI-assisted review, respondents reported a mix of approaches. The leading model was per-document billing (37%), closely followed by hybrid approaches (30%) that combine multiple structures. A sizable portion (21%) said they did not know or considered it not applicable, underscoring the nascent state of GenAI pricing. Far fewer cited per-token (6%), per-GB (4%), or outcome-based pricing (1%).
The dominance of per-document models reflects the industry’s tendency to lean on familiar frameworks when introducing new technologies. Clients are accustomed to measuring review work on a per-document basis, and providers appear to be using that familiarity to introduce GenAI into established billing systems. Yet the significant share of hybrid approaches signals active experimentation, as some providers test ways to integrate token-based billing — common in cloud LLM infrastructure — while still offering per-document simplicity. The market is clearly caught between technical cost alignment and client comfort, with no single dominant structure yet.
Review Pricing - Primary Model for Gen AI-Assisted Review in eDiscovery - Summer 2025
Per-Document Costs: Wide Variability
When it comes to average per-document costs, nearly half of respondents (46%) reported uncertainty or non-applicability, showing how early the market still is. Among defined pricing, the largest group (20%) cited rates of $0.26–$0.50 per document, followed by 17% in the $0.11–$0.25 band. A small but equal share (6% each) reported very low costs (<$0.05 or $0.05–$0.10) or premium pricing above $0.50.
This wide spread underscores a fundamental question: is GenAI meant to reduce costs dramatically, or can it command a premium because of speed and accuracy? Early adopters appear divided. Some are using ultra-low per-document costs to demonstrate efficiency, while others anchor prices closer to traditional review rates to highlight value rather than savings. Over time, consolidation may emerge in the mid-range ($0.25–$0.50), but the current variability illustrates how unsettled the market remains.
Review Pricing - Average Cost Per Document in Per Document Model of Gen AI-Assisted Review - Summer 2025
Per-GB Costs: Rare and Unsettled
Per-gigabyte billing for GenAI-assisted review is not yet widely established. An overwhelming 81% reported uncertainty or non-applicability. Among those with defined rates, the most common tier was less than $25 per GB (7%), followed by $25–$50 (6%). Higher-cost tiers — $51–$100 and above $100 per GB — were cited by just a handful of respondents.
The lack of adoption here reflects a deeper misalignment: per-GB billing, long a staple of processing and hosting, does not map cleanly onto GenAI review. The unit economics of tokens — the actual building block of large language model pricing — don’t correspond directly to data volumes. A single gigabyte of spreadsheets, for example, may generate far more tokens than a gigabyte of PDFs, making per-GB pricing unpredictable and potentially misleading. For this reason, per-GB may ultimately fade from GenAI pricing, supplanted by models that better reflect review outputs, such as per-document or subscription frameworks.
Review Pricing - Average Cost Range Per GB in Per GB Model of Gen AI-Assisted Review - Summer 2025
Outcome-Based Pricing: Rare but Experimenting
Outcome-based models are even less common. Fully 83% of respondents reported uncertainty or non-applicability. Among those exploring such approaches, custom agreements tied to project goals (11%) dominate, while small minorities use performance metrics (4%) or accuracy-based fixed fees (1%).
This hesitation reflects both promise and risk. Clients may be intrigued by the idea of paying for outcomes — accuracy, recall, or efficiency gains — but outcome-based pricing in eDiscovery raises difficult questions. What metrics should define success? How defensible are GenAI results in court or before regulators? And who bears the risk if accuracy targets aren’t met? For now, providers appear cautious, piloting outcome-based models only in bespoke engagements. Longer term, however, such models could become differentiators for vendors confident enough in their technology to stand behind its results.
Review Pricing - Typical Structure of Outcome-Based Pricing Models in Gen AI-Assisted Review - Summer 2025
Handling Complex or Special Documents
Respondents reported no single standard for how GenAI-assisted review addresses complex or special documents. The largest group (39%) cited uncertainty or non-applicability, while 30% said it depends on the specific issue. Another 17% said they revert to manual review at standard rates. Smaller groups reported treating it as additional processing (6%), including it in the base price (6%), or applying per-document surcharges (3%).
This fragmentation highlights one of GenAI’s core challenges: while the technology excels at routine pattern recognition, it struggles with edge cases, such as highly technical files, privileged communications, or foreign-language material. Providers are split between absorbing these complexities into their base pricing, reverting to manual review, or charging add-ons to cover additional work. Over time, as domain-specific LLMs — trained on regulatory filings, contracts, or multilingual corpora — mature, pricing models for complex documents may stabilize. Until then, clients should expect variability and negotiation when GenAI meets difficult content.
Review Pricing - Accounting for Docs That Fail To Process or Require Special Handing (Gen AI) - Summer 2025
Why This Matters
The survey data makes clear that GenAI-assisted review is still in an experimental phase. Per-document and hybrid models dominate today, but wide variability in costs, high uncertainty, and fragmented handling of complex documents reveal a market without benchmarks.
The respondent base also shapes interpretation. With law firms (43%) and service providers (36%) together making up nearly 80% of participants, the findings reflect perspectives grounded in client service and defensibility, not necessarily deep technical cost modeling. And with 90% of responses coming from the U.S., these results largely reflect American pricing dynamics. In Europe, where GDPR and the upcoming EU AI Act impose added compliance obligations, and in Asia-Pacific, where offshore review centers are more established, pricing structures may evolve differently.
Looking Ahead
Generative AI in eDiscovery represents both opportunity and uncertainty.
In the near term, providers will continue to lean on per-document billing and hybrid models because they’re familiar and easy to explain. Clients are likely to favor these structures for predictability, even as they question the fairness of costs that range from fractions of a cent to above traditional review rates.
Over the longer term, as adoption grows and courts weigh in on defensibility, we may see a shift toward value-based and subscription pricing. These models reward efficiency and consistency rather than raw volume. Markets outside the U.S. may leapfrog traditional approaches, adopting AI-first frameworks shaped by local regulatory demands and multilingual needs.
Ultimately, the trajectory of GenAI pricing will hinge less on technology than on trust. If providers can demonstrate that GenAI results are reliable, defensible, and cost-effective, then pricing models will stabilize and standardize. If not, variability and uncertainty will persist, and GenAI may be priced as just another add-on rather than a transformative force.
In Part 7 — the final installment of this series — we will turn to the aggregate results of the Summer 2025 eDiscovery Pricing Survey, drawing together insights across all service areas to provide a holistic view of pricing in today’s market.
“As eDiscovery adapts to rapid technological shifts and mounting regulatory demands, benchmarking pricing is essential,” said Kaylee Walstad, Chief Strategy Officer of EDRM. “ComplexDiscovery’s survey provides the data we need to understand current costs and prepare for the future. EDRM is proud to support this important resource.” – July 9, 2025
Assisted by GAI and LLM Technologies
Additional Reading
- The Human Core of eDiscovery: Review Services in the Summer 2025 Pricing Survey
- Processing, Hosting, and Project Management Pricing: The Engine Room of eDiscovery in the Summer 2025 Survey
- The Front Door of eDiscovery: Forensic Pricing Insights from the Summer 2025 eDiscovery Survey
- The People Behind the Pricing: Respondents to the Summer 2025 eDiscovery Pricing Survey
- Summer 2025 eDiscovery Pricing Trends: Setting the Stage
- ComplexDiscovery OÜ – Winter 2025 eDiscovery Pricing Report: A Market in Transition
- eDiscovery Surveys Archives – ComplexDiscovery
Source: ComplexDiscovery OÜ







































