Government Contracting AI: How CLEATUS Performs
Measured government contracting (GovCon) AI accuracy, usage, and coverage across capture stages, with the sample and method behind each number.
CLEATUS, the agentic AI platform for government contractors, measured the accuracy of its government contracting (GovCon) AI in October 2026. In an AI-scored blind benchmark, more than 98% of sampled findings from its requirements engine were correct and supported by the source. It covered 26% more of the test's checks than the previous engine. More than 95% of the FAR and DFARS clauses it cited appeared in the solicitation text, and it saved the correct value for more than 96% of key fields. This report gives the sample, date, and method behind every number.
To see the coverage behind these numbers, browse the CLEATUS government market, check the set-aside programs that decide eligibility, and use the GovCon glossary for any procurement term in this report.
TL;DR
- More than 98% of sampled requirement findings were correct. In an AI-scored blind benchmark of 40 production solicitation and notice records, more than 98% of sampled findings from the CLEATUS requirements engine were correct and supported by the source. The engine also covered 26% more of the test's strict checks than the previous one.
- More than 95% of cited FAR clauses trace to the solicitation. In a 500-contract audit of more than 47,000 FAR and DFARS clause citations, more than 95% appeared in the solicitation text.
- Key fields were correct more than 96% of the time. Across 13 fields on 75 held-out contracts, CLEATUS saved the correct value for more than 96% of field values, up from 94% for the previous version.
- State and local portals hold more open bids than SAM.gov. On October 22, 2026, state and local portals held 46% of the open opportunities CLEATUS tracked, and SAM.gov held 32%.
- Usage is growing. CLEATUS created more than 2,200 AI Proposal Writer sessions in the 90 days ending October 22, 2026, up 27%, and nearly 58,000 AI Workflow runs completed in the 90 days ending October 23, 2026, up 31%.
Why AI accuracy in government contracting is hard to judge
Government contractors now have dozens of AI tools to choose from, and few of them publish measured results with the method behind them. Public comparisons in our market often use short vendor-run search tests. Buyers are left to run one real pursuit through each tool and judge the output by eye.
That is a problem because the errors that matter in GovCon are specific and expensive. A missed instruction in Section L, the instructions to offerors defined in FAR 15.204-1, can make a proposal noncompliant before an evaluator reads the technical volume. A deadline extracted without its time zone can cost a submission by an hour. A wrong set-aside can send a company after a bid it is not eligible to win, and an invented FAR clause number puts a false requirement into a compliance matrix. Real solicitations are multi-document packages with attachments, exhibits, and amendments, so each of these errors has many places to hide.
General-purpose AI benchmarks don't help much here either. They measure reasoning and coding, not whether a model can read a 300-page DLA package and find the packaging requirement in Attachment 4. Industries like law and finance have started publishing domain benchmarks. Government contracting has fewer widely used benchmarks, so a contractor evaluating AI has no common yardstick.
What a vendor can do is publish its numbers, explain how it got them, and let buyers check. Every figure below names its sample, date, and source, and the methods section describes each test.
CLEATUS GovCon AI performance by stage
The numbers below come from CLEATUS production data and from tests with fixed answer keys. Each one states its sample and its date.
Find: coverage beyond SAM.gov
CLEATUS tracks federal, state, local, and education buyers, not one portal. On October 22, 2026, the public CLEATUS government directories held these records:
| Coverage | Count |
|---|---|
| Public agencies tracked | 62,000+ |
| States covered | 50 + DC |
| Counties covered | 1,600+ |
| Opportunities tracked | 8M+ |
| National stock numbers | 17M+ |
On the same day, more than two-thirds of the open opportunities CLEATUS tracked came from sources other than SAM.gov. State and local portals alone held 46%, against 32% for SAM.gov.
| Measure | Share of open opportunities |
|---|---|
| State and local (SLED) | 45.8% |
| SAM.gov | 32.0% |
| DIBBS | 9.9% |
| Procurement forecasts | 5.0% |
| International | 4.3% |
| Grants | 1.5% |
| GSA and other sources | 0.8% |
| SBIR and STTR | 0.6% |
Open opportunities in CLEATUS on October 22, 2026. Open means active, not awarded, and with a response deadline on or after that date, or none stated.
A contractor who watches only SAM.gov sees about one open opportunity in three. That gap matters most for firms that sell to cities, counties, school districts, and state agencies, where SLED contracts are posted on separate portals with their own registration rules. It also matters for product suppliers, because DIBBS requests for quote move on short clocks and are covered in our DIBBS guide. Auto Capture is the CLEATUS feature that pulls these sources into one feed and matches them to a capture profile.
Monitor: amendments
Amendments change due dates, quantities, and evaluation criteria, and a missed amendment can turn a compliant proposal into a noncompliant one. Between August 28 and October 23, 2026, CLEATUS detected more than 75,000 amendment notices, each linked to the opportunity it changes. That link lets a capture team see the change on the pursuit it's already working, instead of finding it on a portal later.
Recommend: matching at scale
In the week of October 15 to 22, 2026, CLEATUS completed more than 360,000 opportunity-scoring runs against company capture profiles.
Teams act on those matches. In the 90 days ending October 22, 2026, nearly 14,000 CLEATUS recommendations were added to pipelines after scoring. Pipeline acceptance is a customer-validated relevance signal. It does not equal statistical precision, because a team can pass on a good match for reasons of capacity or timing. The published customer results point the same way: D2 Government Solutions reports 75% faster opportunity discovery with the same team.
Shred: requirements, FAR clauses, and key facts
CLEATUS breaks down opportunities before a user asks. Shredding errors cost the most, so this stage gets our strictest tests.
Requirements. We built an AI-scored blind benchmark from 40 production solicitation and notice records that had never been used to develop the system. The set holds 10 DIBBS requests for quote, 8 state and local solicitations, 10 SAM.gov solicitations and combined synopses, 5 sources sought notices, 3 presolicitation or special notices, 2 DIBBS awards, and 2 from other sources. Before any system ran, 478 checks were written from the source documents alone, each one a consequential thing a contractor must know or do to respond.
In an AI-scored random sample of 200 findings per engine, more than 98% of the current production engine's findings were correct and supported by the source, up from about 95% for the previous engine. The current engine also covered 26% more of the test's checks and scored higher on 30 of the 40 records.
| Measure | Previous version | Current version |
|---|---|---|
| Requirement findings correct and supported | 95.5% | 98.5% |
| Key fields correct | 94.0% | 96.8% |
Requirements: AI-scored blind benchmark of 40 production solicitation and notice records, 478 checks, and 200 sampled findings per engine, October 19, 2026. Key fields: 13 fields on 75 held-out contracts, three runs each, October 16, 2026.
The checks are strict: each one asks for a specific date, quantity, address, form, or condition, not a general topic. AI reviewers write and grade the checks, so the benchmark is a reproducible AI-scored comparison, not an independent human audit. We hold the test fixed so that every future version is measured the same way.
Requirement quotes trace back to the source. In our sample of 50 live public opportunities on October 22, 2026, every saved requirement carried a quote from its source document. We searched each quote in the available solicitation text, and more than 95% of the 800+ quotes appeared in the source after spacing and punctuation were normalized. This live quote audit is a separate test from the blind benchmark above.
FAR clauses trace back to the source. An invented clause number is one of the most damaging errors a compliance tool can make. We audited 500 public contracts with a CLEATUS FAR breakdown, taking up to 50 from each source group where available and filling the remaining sample by recency. The audit covered more than 47,000 clause citations. More than 95% of the cited FAR and DFARS clauses appeared in the solicitation text. This source-text check does not classify agency-specific clauses against the central eCFR list. Readers can check any clause against FAR Part 52 and DFARS Part 252.
Key facts. CLEATUS also extracts the fields that drive go/no-go decisions. They include response deadline and time zone, notice type, response type, set-aside type, solicitation number, contract vehicle, estimated value range, period of performance, estimated award date, and product or service type. Across 13 fields on 75 held-out contracts, run three times each, CLEATUS saved the correct value for more than 96% of field values, up from 94% for the previous version.
This is the work Contract Breakdown does when an opportunity arrives, and it's why a capture manager can start a bid/no-bid decision from a structured summary instead of a stack of PDFs.
Market intelligence: competitors
Knowing who is likely to bid changes how you price, team, and position. On October 22, 2026, CLEATUS market intelligence drew on nearly 148,000 competitor records. Market Intelligence is the CLEATUS feature that brings these records together with agency hierarchies, contracting officers, and competitor award history.
Write: proposals in the AI Proposal Writer
Use of the AI Proposal Writer is growing. CLEATUS created more than 2,200 proposal sessions in the 90 days ending October 22, 2026, up 27% from the previous 90 days. In the 90 days ending October 23, 2026, CLEATUS agents authored more than 92% of the 30,000+ document versions saved in proposal workspaces. That's a file-version share, not a measure of document length or drafting time.
The published customer results show what that means for a proposal team. MST Maritime Management went from three proposals a month to more than ten. D2 Government Solutions reports 80% less draft development time and three times the proposal output with the same team.
Automate: AI Workflows
AI Workflows run recurring capture and business development jobs, such as intake triage, document shredding, and pipeline reporting, in the background. In the 90 days ending October 23, 2026, nearly 58,000 workflow runs completed, up 31% from about 44,000 in the previous 90 days.
| Measure | Previous 90 days | Last 90 days |
|---|---|---|
| AI Proposal Writer sessions created | 1.8K | 2.2K |
| AI Workflow runs completed | 44.2K | 57.8K |
In thousands. Each row has its own scale. Proposal sessions: 90 days ending October 22, 2026. Workflow runs: 90 days ending October 23, 2026.
Platform: tools, connectors, and scale
The CLEATUS platform has more than 170 agent tools. Tools are what let an agent do work instead of describing it: search opportunities, pull an award history, read a solicitation, create a saved search, or start a workflow. The tools available in a given place depend on the surface, so the GovCon Copilot in the dashboard and the connectors for Claude and ChatGPT expose different sets.
Over the 14 days ending October 22, 2026, CLEATUS product agents processed more than 30 billion tokens. That is a two-week measure, separate from the cumulative milestone in our 10 billion token announcement.
The outcome measure we publish separately is customer wins. Under a dated rule explained in how CLEATUS customers recorded more than $5B in government contract wins, CLEATUS customers had recorded more than $5B in government contract wins as of September 8, 2026.
How we measured government contracting AI accuracy
Production data. Coverage, amendment, recommendation, market intelligence, proposal, workflow, and token figures come from CLEATUS production records read on October 22 and 23, 2026. Each figure states its own date range. Counts are totals across all accounts, and no customer, company, or document is identified.
Requirements test. The 40 production solicitation and notice records were never used in development and span the main notice types. The 478 checks were written from the source documents before any version ran, and findings were graded against the source without knowing which version produced them. AI reviewers performed both steps, and no human spot check was completed.
Key-field test. AI reviewers labeled 100 production contracts field by field against the source documents. 75 of them were held out from all tuning, and each version ran three times on each held-out contract.
Quote and FAR checks. Quotes were checked on 50 live public opportunities, and FAR citations on 500 public contracts, both drawn across source groups. Each quote and clause number was searched in the linked document text available for its record.
Pipeline acceptance. Pipeline acceptance counts recommendations that were added to a pipeline after CLEATUS scored them. The event is a customer action signal, not an independent correctness label.
What these numbers are not. The tests measure a fixed sample, not every opportunity. We rerun them when the system changes and update this page with the new date.
Frequently asked questions
The fastest way to judge these numbers is to test them on your own work. Connect a capture profile, pick a solicitation you've already shredded by hand, and compare the CLEATUS breakdown with yours.
Further Reading
- 10 AI Prompts to Decode Any Government Solicitation in 30 Minutes
- AI-Driven Bid/No-Bid Decisions
- How to Win SLED Contracts
- DLA DIBBS: Login and How to Win DLA Contracts
- How CLEATUS Customers Recorded More Than $5B in Government Contract Wins
Customer Stories
- How D2 Government Solutions Tripled Growth Without Adding Staff
- How MST Maritime Quadrupled Proposal Output with CLEATUS AI
- How Operation Hired Achieved 6× Proposal Output with CLEATUS AI
- How Cuadrant Won a California Air National Guard Contract in Under 3 Hours
About CLEATUS
CLEATUS is an agentic AI platform that helps government contractors discover the right opportunities, manage capture pipelines, and write winning proposals. Automate your GovCon operations end-to-end: build custom AI-powered automations that handle multi-step processes on autopilot, a force multiplier for your capture and BD team that works in the background while you stay in the loop on every decision. We aggregate federal, state, local, and city opportunities; our GovCon Copilot analyzes solicitations and your internal documents to deliver actionable market intelligence that drives revenue growth.
