The AI Examiner — legal discovery document review, 2026-08-08

Headline

The only publicly verified price for a fully reviewed discovery document is a 2020 federal court order — $754,029.46, or roughly $2.34 a document — five to twenty times above what vendors advertise for AI-assisted review in 2026.

Summary

Every other figure in this market comes from a vendor's rate card or a buyer survey, and both have a side to take. The one number a court actually checked line by line is thirteen years newer than the last independent attempt to measure the whole discovery pipeline, and it predates the generative-AI review tools now priced as low as eleven cents a document. Two market-level indicators point toward why a cheaper, equally verified figure hasn't surfaced to replace it: the share of discovery spending that goes to review has fallen only modestly since 2012 even as data volumes have grown roughly five times faster than total spending, and a survey of legal buyers finds most law firms hold rates flat or raise them when AI cuts review time, rather than billing less.

The evidence

The only audited price on record

In 2012, the RAND Corporation studied 57 large-volume discovery productions from eight Fortune 200 companies and found that review — checking documents for relevance, responsiveness, and privilege — consumed 73 cents of every discovery dollar, at a cost of roughly $14,000 per gigabyte reviewed, against $20,000 per gigabyte for the full production (fetched, observed — see Gaps on sourcing confidence for this figure). No comparably designed, independently run study of actual discovery spend has been published since.

The closest thing to a modern equivalent is not a study but a court order. In Lawson v. Spirit Aerosystems, Inc. (D. Kan., Magistrate Judge Mitchell, affirmed by District Judge Melgren, Dec. 2020), the court shifted $754,029.46 to the plaintiff for a technology-assisted review the plaintiff had insisted on: $449,999.21 paid to the vendor, Legility, plus $304,030.25 in attorneys' fees, for a review of 322,000 documents of which only 3.3% turned out to be responsive (fetched, observed — the dollar figures are corroborated across four independent legal-industry case summaries citing the same order). Divided across the full document set, that is roughly $2.34 per document, all in. A second case, In re Biomet (N.D. Ind. 2013), shows the same pattern at larger scale: the defendant had already spent about $1.07 million reviewing part of a 19.5-million-document collection, with total costs projected between $2 million and $3.25 million, when the court rejected a plaintiff request to redo the review with predictive coding (fetched, observed).

Both figures come from disputes, not routine matters — cost figures enter the public record through litigation over discovery, specifically because one side thought the other's approach was disproportionate. That is a structural feature of this source type, not a limitation of this search: courts do not audit discovery costs unless someone asks them to, and nobody asks in a matter that went smoothly. It means the only independently verified per-document prices in the public record are drawn from the worst-behaved cases, not typical ones — and, as of this run, none of them is newer than 2020.

What vendors advertise now

A survey of 53 eDiscovery providers and buyers, run by ComplexDiscovery in partnership with EDRM between late December 2025 and February 21, 2026, found generative-AI-assisted review pricing clustering at $0.26–$0.50 per document (20.8% of respondents), with meaningful shares also reporting $0.11–$0.25 (15.1%) and $0.05–$0.10 (15.1%) (fetched, observed). Per-gigabyte pricing was harder to pin down: 64.2% of the same respondents answered "do not know / not applicable" when asked their per-GB rate, a level of uncertainty the survey itself flags as notable. Separately, industry coverage puts human review at $1.50–$3.00 per document and earlier technology-assisted review at $0.50–$1.50 per document "as recently as two years ago" (fetched, self-reported/inferred — an industry blog with its own commercial interest in the AI-pricing narrative; collection rates it cites, $250–$350/hour, and expert-testimony rates above $550/hour, were not independently corroborated this run).

Set against Lawson's $2.34-per-document, court-verified figure, today's advertised AI rates are five to twenty times lower. No vendor or buyer in this market publishes an audited, per-matter figure showing that gap actually closing on a real invoice — only rate-card numbers on one side and a five-year-old litigated dispute on the other.

Why review's share of the bill hasn't moved to match

EDRM's own market modeling puts worldwide eDiscovery spending at $4.73 billion in 2012, rising to roughly $19.61 billion in 2025 and a projected $28.08 billion by 2030 (fetched, observed). Within that spending, review's share was 73% in 2012 — the RAND figure — falling to 62% by 2025 and a projected 52% by 2030 (fetched; the 2012 figure is observed, but EDRM describes the 2025 figure as "reconciled modeling," not a direct measurement, so it is carried as inferred). An 11-point drop over thirteen years, while advertised per-document review prices fell several-fold in the last two years alone, is a smaller shift than the sticker-price collapse would predict — unless something else in the pipeline grew to absorb the difference.

EDRM names that something directly: enterprise data volume is growing at roughly 35% a year, against roughly 7.44% annual growth in total eDiscovery spending through 2030 (fetched, observed — EDRM's own stated figures). Reviewing a document has gotten radically cheaper, but there are radically more documents to review — chat platforms, video, and cloud communications that a 2012 discovery matter would not have touched. EDRM states the volume growth; it does not connect that growth to whether a client's bill for a given matter actually falls. That connection — a multi-fold price drop showing up as only a low-double-digit percentage shift in spending mix, over a period in which the thing being priced also multiplied in volume — is not stated by any single source found this run.

Where the difference is going instead

A survey of more than 600 senior legal leaders across eight countries, run by Axiom, found that only 6% of law firms reduce fees when AI cuts the time a task takes; 34% charge premium rates for AI-assisted work instead, and the remainder hold rates flat. Ninety percent of legal spending still moves through standard hourly billing. Document review is named specifically among the AI-accelerated tasks the survey covers, alongside legal research, contract analysis, and brief writing (fetched, self-reported — Axiom is an alternative legal-services provider that competes with the law firms this finding criticizes, a real motive to publish it; the survey's own methodology disclosure omits dates and respondent titles).

Vendors show a parallel pattern. In October 2025, Relativity announced that its aiR for Review and aiR for Privilege tools would be included in standard RelativityOne subscriptions at no additional charge starting in early 2026 — but the mechanism was a restructuring to "an all-inclusive, unlimited fixed-fee per-GB structure" replacing the prior volume-metered add-on, not a cut to the underlying per-GB subscription rate itself (fetched, observed — Relativity's own announcement). DISCO is reported to have made a similar move, collapsing its platform into a single per-GB fee (fetched, inferred — reported secondhand in industry pricing coverage, not confirmed against a DISCO announcement directly). In both cases, the AI got cheaper without a corresponding public claim that the total bill did.

On the buyer side, in-house counsel report the same gap from the other direction. Legal departments increased AI spending in 2026 while pressing outside counsel harder for cost reductions — pressure that only makes sense if the reductions have not been showing up on their own (fetched, observed, Deloitte survey as covered by Law.com). A separate April 2026 survey found a large majority of in-house attorneys had yet to see cost savings from outside counsel's use of generative AI (fetched, inferred — figure not independently re-fetched; exact percentage and methodology are as characterized in secondary coverage). Thomson Reuters reported law firm billing rates rose 7.4% year over year in the second quarter of 2025, against roughly 2.8% U.S. inflation, with Am Law 100 lawyers increasingly crossing $1,000 an hour while the broader market averages around $600 (fetched, observed).

What people are asking

The public debate over this question runs almost entirely through buyer-side and trade-press channels rather than open practitioner forums. On the buyer side, legal-tech commentary is explicitly framing the question as "who gets the AI dividend" — whether efficiency gains from legal AI flow to the client or disappear into law firm margin — and arguing that clients should not assume they get it by default (fetched, observed, Artificial Lawyer, July 2026). On the labor side, entry-level legal hiring has been declining, paralegal roles are described as under direct, measurable pressure, and a remote contract-attorney document-review position was found advertising a starting rate of $23.00 an hour in 2026 (fetched, observed — a live job posting). No dedicated open-forum thread (Reddit, Slack, or similar) specifically debating discovery-review pricing was found this run; the conversation appears to live in surveys, legal-tech trade coverage, and law-firm client alerts instead.

What this means for a litigation support manager renewing a discovery vendor contract

A vendor's advertised per-document AI rate is not the number that determines the size of next year's invoice. The evidence here points to the per-GB base or subscription fee — the part vendors have restructured rather than cut when they added AI at "no additional charge" — as the actual lever on total cost. Nothing in the market data or the buyer surveys found this run shows the recent price compression arriving on a bill without that fee being separately negotiated down.

Methodology & sample

This issue rests on roughly twenty WebSearch queries and a dozen WebFetch pulls, all fetched-tier; examiner/sources/ held no files, so there is no operator-supplied data here. Confidence is strongest on the two court cases (multiply corroborated, primary-adjacent) and the Winter 2026 ComplexDiscovery/EDRM pricing survey (n=53, directly stated figures). It is weaker on the RAND 2012 figures, which could not be re-fetched from RAND's own site this run and rest on converging secondary citations. It is weakest on the Axiom survey (undisclosed methodology, a competitor's commercial interest in the finding) and on the GC savings-survey and DISCO-pricing claims, both of which came through search-tool summaries rather than a direct primary fetch.

Gaps

• No Canada-specific pricing, cost-shifting, or survey data was found for this focus despite a direct search; every figure in this issue is US-sourced, and this issue covers the US market only.
• RAND's own pages (rand.org) returned an expired certificate and a 403 error on direct fetch; the $14,000/$20,000-per-gigabyte figures rest on two secondary citations that agree with each other, against a third that cites $18,000 instead — not resolved.
• EDRM's 2025 figure for review's share of discovery spend (62%) is described by EDRM itself as "reconciled modeling" bridging the 2012 RAND figure and a 2030 projection, not a direct 2025 measurement.
• The Axiom survey (6%/34%/58% law-firm billing behavior) is the single largest data point behind the capture argument in this issue, and it comes from a vendor that competes commercially with the law firms it critiques; its methodology section discloses no survey dates or respondent breakdown.
• The molawyersmedia GC-savings-survey figure and the report of DISCO's pricing restructuring could not be independently re-fetched this run (403 error and secondhand sourcing, respectively) and are carried at lower confidence than the rest of this issue.
• No comparably designed, independently run successor to RAND's 2012 study was found. That gap is itself treated as a finding in the body (see "The only audited price on record"), not restated here.
• Demand-survey coverage for this focus came from trade press, buyer surveys, and legal-tech commentary rather than an open practitioner forum; no Reddit, Slack, or similar community thread specifically on discovery-review pricing was located this run.