Skip to content
Consight logo

AI for Construction Estimating: How GCs Use It Today

Photo of Chris ToubiaChris ToubiaPreconstruction insights

Commercial general contractors use AI for estimating in three jobs. Takeoff tools measure quantities from drawings. Document tools read plans and specs and draft scopes of work. Bid leveling tools read subcontractor proposals and line them up against your scope. In each job the AI does the first read or the first count. An estimator still checks the result, prices the work, and owns the number.

Adoption is broad but early. In the 2026 Construction Hiring and Business Outlook from AGC and Sage, 61% of 951 responding firms said they use AI or plan to increase investment in it. The same survey found firms use AI most for office and administrative work, estimating, and preconstruction.

Where AI fits in a GC's estimating workflow

Each estimating task calls for a different kind of AI tool. The table shows what the AI does in each task, what stays with you, and examples named from each vendor's own pages.

Estimating task What the AI does What the estimator still does Examples
Quantity takeoff Detects and measures areas, lengths, and counts on drawings Checks scale and conditions, and adds work the drawings do not show Togal.AI, STACK AI Takeoff, Kreo
Plans and specs Reads the documents and drafts a scope of work for each trade Edits each scope and decides which trade carries the interface items ChatGPT, Claude, Consight's Generate scope
Subcontractor bids Reads each proposal and marks each scope line included, excluded, or unclear Reviews the calls, adds plugs for missing scope, and picks the award Procore's Bid Leveling Agent, Consight
Unit costs and markup Suggests numbers, but a general model has no view of your cost history Sets prices from your own history and this bid's sub quotes Your estimate and cost database

Togal.AI describes its takeoff tool as one that "automatically detects, measures, and compares directly from your drawings." Procore says its Bid Leveling Agent "compares sub proposals against specs and drawings in a standardized matrix." Consight, which publishes this guide, is bid leveling and bid management software.

How much of the AI's bid reading do estimators review?

In Consight's own bid data, estimators review part of the AI's coverage calls and change few of the calls they review. A coverage call is the mark on one scope line for one bidder: included, excluded, or unclear. The figures below come from Consight's bid benchmark snapshot dated November 2, 2026.

  • Calls reviewed: estimators had confirmed or changed 38% of line-item coverage calls. The rest are AI calls no one has reviewed yet. Without the largest organization, which holds 30% of the sample, the share is 33%. This is a recorded figure over 5,586 bids, 1,873 packages, 195 projects, and 66 organizations.
  • Calls changed: of the calls estimators reviewed, they changed 5% rather than confirmed them. Without the largest organization, which holds 39% of that sample, the share is 3%. This is an estimator-reviewed figure over 2,609 bids, 956 packages, 150 projects, and 60 organizations. A change may correct the call itself or only its price or note.

Read the second figure with care. Estimators choose which calls to review, so 5% is not an error rate for every call the AI makes. The figures do show how the work splits. The AI makes most of the first calls, and estimators review the calls they choose to check.

Can ChatGPT or Claude help with construction estimating?

ChatGPT and Claude help with the reading and writing parts of estimating. They cannot give you quantities or prices you should bid on. Both are general chat models. They read what you paste or upload and write a fluent answer, but they know nothing about your project beyond that.

Reading a plan set with Claude's API cost about $9 a pass in one 1,200-page example, before the document handling and bid comparison a production workflow adds.

What chat models do well for estimators

  • Summarize a long proposal or spec section into a list of inclusions, exclusions, and qualifications.
  • Draft a scope of work narrative from your notes, for you to edit.
  • Build a checklist, such as bid day steps or questions for a pre-bid meeting.
  • Draft RFIs, and emails asking a bidder to confirm or price a gap.
  • Set two short exclusion lists side by side so you can spot the differences.

Where chat models fail at estimating

  • Quantities. A chat model does not give you a scaled, measured takeoff you can check line by line. Unless you supply the quantities, any it states are guesses.
  • Unit costs. A chat model has no view of your cost history or this week's sub quotes. Ask it for a price per square foot and it returns a confident number with no source behind it.
  • Source tracing. A chat answer does not link each figure to the page of the proposal it came from. You have to ask for quotes and then check each quote yourself.
  • Confidential bid documents. Sub pricing is confidential. Check how the tool treats uploads before you paste in a sub's proposal.

Anthropic's guide to reducing hallucinations warns that models "can sometimes generate text that is factually incorrect or inconsistent with the given context." The same guide suggests three fixes that suit bid review. Let the model say it does not know. Have it pull word-for-word quotes before it answers. Have it drop any claim it cannot back with a quote.

A prompt that keeps a chat model honest on a proposal

Ask for quotes, not conclusions. This prompt works for a first read of one subcontractor proposal:

Using only the attached proposal, list every exclusion, qualification, and allowance word for word, with the page it appears on. If the proposal does not mention an item, write "Not stated." Do not estimate any price or quantity.

The answer is still a first read. Check each quote against the proposal before you carry a plug or call the bidder.

What ChatGPT and Claude do with your uploads

Each vendor sets its own default for training on what you upload. On personal ChatGPT Free, Plus, and Pro accounts, OpenAI's data controls guide says data sharing is on by default. You can turn off "Improve the model for everyone."

OpenAI says it does not train on content from ChatGPT Business, Enterprise, and Edu by default. On Claude's consumer plans, you choose whether Anthropic uses your chats for training. Check your company's AI policy before you upload a sub's pricing.

Are there free AI tools for construction estimating?

A few AI tools have a free option, but the free tiers leave out most of the AI. The general chat models have free plans with usage limits. Most estimating and takeoff vendors offer a free trial or a demo instead. Each row below comes from the vendor's own pricing page, checked on November 2, 2026.

Tool What is free Limits to know
ChatGPT A Free personal plan Data sharing for training is on by default on Free, Plus, and Pro
Claude A $0 Free plan Limited usage every five hours and every week
STACK A free plan after 7 days of Pro: 1 active project a month, up to 300 plan pages, 50 measurements per project AI Takeoffs and STACK IQ start in Premium, at $299 per user a month, billed annually
Kreo A 7-day free trial of the Essential, Pro, and Business plans No free plan; the Essential plan has no AI takeoff
Togal.AI No free plan or trial on its pricing page; access starts with a demo The Growth plan is $299 per user a month, billed yearly
Consight A 14-day free trial with full access and no credit card Built for bid leveling; no free plan after the trial

Sources: Claude pricing, STACK pricing, Kreo pricing, Togal.AI pricing, and Consight's pricing.

Can AI do construction estimates for free?

Free AI can draft parts of an estimate, but it cannot produce a bid-ready estimate on its own. A free chat plan can summarize specs, draft scope language, and build checklists. A free takeoff plan can measure a small job by hand, with the AI features held back for paid plans.

The quantities, unit costs, and subcontractor coverage that decide the bid still come from your own takeoff, your cost history, and your review of each proposal.

How Consight uses AI in preconstruction

Consight uses AI to read subcontractor bids and plan documents. Consight leaves pricing to you. Each step below matches the help center:

  1. Generate scope. Upload plans and specs, build bid packages in BidGrid, and click Generate scope to draft scope sheets. See build bid packages from plans or a takeoff.
  2. Bring in a takeoff. To start from quantities, upload a takeoff from Togal.AI, Bluebeam, or OST as a bid form or takeoff.
  3. Level Bids. Consight reads each bid and fills each bidder's column on the scope sheet as Included, Not Included, or Needs Review. Matching usually finishes in about a minute. See what Level Bids does.
  4. Check the source. View in Bid jumps to the exact spot in the proposal a match came from. See why you should still review AI matches and how to trace bid leveling results to the source.
  5. Plug the gaps. Connie, the assistant in Consight, will not invent a price for a plug. She asks you for one. See add a plug for missing or unclear scope.

Consight does not train on customer files, and other companies never see your project data. See how your data is protected.

For how AI bid leveling works step by step, read AI bid leveling explained. To test a tool on your own bids, read how to pilot AI bid leveling software. To learn where your team would use AI first, start with the AI readiness survey for estimating teams. For measured run times across 460 runs, read Consight's AI performance benchmarks.

How we measured

Both Consight figures come from the bid benchmark snapshot dated November 2, 2026, computed read-only from Consight's production bid data. The snapshot leaves out Consight's own organizations, deleted records, archived bids, and test, demo, sample, and training projects.

The snapshot suppresses any figure from fewer than 5 organizations or fewer than 30 packages. Each figure also has a value without the organization that contributes the most observations. That organization holds 25% or more of both samples here, so this guide gives both values. The snapshot has no trend over time, names no customer, and publishes no single bid's amount.

Where Consight fits

Consight is the preconstruction workspace built around this kind of work—from understanding project documents and building scopes to collecting bids, leveling proposals, reviewing risk, and documenting the final award.

Put it into practice on your next project.

Talk to sales

Trust your bid on bid day. Get Consight.

Start now