Engineering and architecture
AI in geotechnical engineering: what firms actually use it for in 2026
By SourceX Editorial · Updated
Short answer
AI in geotechnical engineering is used mostly for document work in 2026: extracting data from old boring logs, checking lab results, drafting report sections and searching past projects. Predictive design tools remain largely research and pilots. The working rule: AI can prepare inputs and drafts, but a licensed engineer owns every interpretation and recommendation.
Key takeaways
- The most practical geotechnical AI uses today are extraction, search and drafting, not design.
- Boring logs, lab test results and past reports are the inputs that decide whether AI tools work.
- Records linking an investigation to measured field performance, such as settlement monitoring, are rare and valuable.
- Recommendations, bearing values and sealed reports stay with the engineer of record.
- A firm's archive can be both an internal AI input and, after rights review, a licensable asset.
What do geotechnical firms actually use AI for?#
Geotechnical firms use AI mainly to handle documents: pulling depths, blow counts and soil descriptions out of scanned boring logs, organizing lab data, drafting standard report sections and finding relevant past projects near a new site. These tasks are repetitive, well defined and easy for an engineer to check against a source.
Predictive uses, such as estimating soil parameters from nearby borings or forecasting settlement, get attention in research journals and vendor demos. In practice they serve as screening aids at most, because site variability, sparse data and professional liability limit how far a firm can rely on them.
The bottleneck is usually the firm's own records. Many sit in PDFs, legacy gINT projects and paper files that no tool can use until someone gives them structure.
Use cases by records, maturity and review#
Each geotechnical AI use case depends on a specific set of firm records, and the review it needs rises as the output moves closer to a design decision. Maturity ratings below are qualitative and describe typical practice, not any one vendor.
| AI use case | Firm records it needs | Maturity | Human review required |
|---|---|---|---|
| Boring log extraction | Scanned or digital boring logs, log legends, project coordinates | In routine use at some firms | QA sample checked against source logs |
| Lab data checks | Moisture content, Atterberg limits, gradation, consolidation and strength results | Practical with structured lab data | Lab manager reviews flagged values |
| Report drafting | Past reports, standard language, field and lab summaries | Practical for boilerplate and summaries | Engineer of record edits and approves |
| Project search near a site | Report library indexed by location, project type and soil conditions | Practical once files are indexed | Engineer judges relevance |
| Subsurface interpolation and stratigraphy | Boring and CPT data with coordinates and elevations | Pilots and specialist tools | Engineer validates against borings |
| Parameter or settlement prediction | Lab and field data linked to measured performance | Mostly research | Screening only; design uses accepted methods |
- 01Your work records
- 02Personal details removed
- 03You approve the release
- 04Licensed to AI developers
Why boring logs come first#
Boring logs come first because nearly every other geotechnical AI use depends on them. A log ties a location and elevation to depth intervals, sampler data, blow counts, soil classifications, groundwater observations and references to lab tests.
Many firms hold many years of logs in mixed forms: typed logs on paper, scans of those logs, exports from gINT or similar software, and PDFs embedded in final reports. Structured formats such as AGS and DIGGS give firms a target schema, so extracted data can move between tools without rework.
Firms that convert logs once into a consistent schema get repeated benefit: faster desk studies, better-informed proposals and a cleaner base for any future model.
Where AI must not replace engineering judgment#
AI must not replace engineering judgment wherever an output becomes a recommendation a client or contractor relies on. Bearing values, pile capacities, slope stability conclusions and earthwork recommendations carry professional responsibility that a tool cannot hold.
- Never let a model fill gaps in a boring log; missing data stays marked as missing.
- Record a reviewer's name and date on every AI-assisted extraction and draft.
- Treat predicted parameters as screening values until tested or checked against accepted methods.
- Check professional liability policy terms and client contracts for AI restrictions or disclosure duties.
- Keep the source document linked to every extracted value so an engineer can trace it.
Which records make geotechnical AI worth the effort?#
The records that make geotechnical AI worth the effort connect investigation data to what happened during and after construction. Most archives hold only the first half of that story: borings, lab tests and the report.
The second half is rarer: construction observation reports, pile driving records, load test results, inclinometer and piezometer readings, and settlement monitoring. A firm that kept these alongside its investigations holds linked evidence that is hard to find anywhere else, useful for its own tools and of interest to outside AI developers.
Start by tagging those projects in the archive index rather than digitizing everything. A short list of jobs with investigation, construction and monitoring records in one place is more useful to an internal pilot, and easier to review for rights, than a large pile of unlinked reports.
Is a geotechnical archive an asset beyond internal AI?#
A geotechnical archive can be an asset beyond internal AI, because model developers need expert, real-world engineering records that are not on the public web. Whether a firm can license specific records depends on contracts, confidentiality and who provided the data.
Some editions of the EJCDC E-500 owner-engineer agreement, for example, say documents are instruments of service and the engineer retains an ownership and property interest, including the copyright and the right of reuse at the engineer's discretion. Wording differs by edition and negotiated changes, and many firms work under owner-drafted contracts instead, so each project's signed terms are reviewed with counsel.
| Record type | Typical rights question |
|---|---|
| Boring logs and lab results the firm produced | Do client contracts restrict reuse or require confidentiality? |
| Reports and recommendations | Does the agreement treat them as instruments of service the firm retains? |
| Owner-furnished data and prior reports | Usually belong to the owner or another consultant and are excluded |
| Public agency projects | Agency terms and public records rules may apply |
| Monitoring data | Who commissioned the monitoring and who controls the readings? |
Illustrative: a geotechnical firm sorts its archive#
Illustrative: a fictional geotechnical and materials testing firm in the Southeast holds boring logs in gINT projects, scanned PDFs and paper files, plus final reports on a file server organized by year. The principal wants AI-assisted desk studies for proposals.
The firm first extracts header and interval data into one schema, linking each value to its source page, and proposal teams can now search nearby borings by location. During the work the data manager flags projects with long-term settlement monitoring and pile records. The principal asks for a rights review on those, because the investigation-to-performance record is the part of the archive outsiders cannot reproduce.
How SourceX approaches geotechnical records#
SourceX treats a geotechnical archive like any operational record set, through the SourceX five-step transaction: Supply, Rights, Preparation, Approval and Delivery. The fit check uses metadata such as record types, formats, years covered and contract forms, and nothing is shared at that stage.
Owner-furnished data, export-controlled work and projects with strict confidentiality are carved out. For anything that proceeds, a SourceX Evidence Packet records provenance, licensing rights, permitted use, the privacy record and release authorization. The firm keeps ownership; records are licensed, not sold.
Frequently asked questions
Can AI read handwritten field logs?
Sometimes, with mixed accuracy. Handwritten logs, faded carbon copies and stamped corrections cause the most errors. Most firms get better results by rescanning the worst pages, extracting typed logs first and routing handwritten ones to manual entry or heavier review.
Will AI tools replace gINT or similar logging software?
Not in the near term. Logging software remains where firms create and manage boring logs to a set format. AI tools sit around it, extracting data from older logs, checking entries and drafting summaries that an engineer then reviews.
Do clients need to know if AI helped prepare a report?
Check the contract first. Some clients now include AI clauses that require disclosure or limit tools that process their data. Even where no clause applies, firms often describe their review process so clients understand that a licensed engineer checked and approved the work.
Is CPT data more useful for AI than SPT boring data?
CPT data is continuous and digital, which makes it easier to process, while SPT data is often sparse and varies across drillers and hammers. Both are valuable when they carry coordinates, elevations and equipment details, and especially when paired with lab results from nearby borings.
How should a geotechnical firm start with AI?
Start with one document task an engineer can check quickly, such as extracting boring logs for a single active site or summarizing past reports within a set radius. Measure accuracy against the source, write down the review steps, and expand only once the results hold up.
Sources
- EJCDC E-500 Owner-Engineer agreement language states that all Documents are instruments of service and the Engineer retains an ownership and property interest therein, including the copyright and the right of reuse at the Engineer's discretion. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.