ChatGPT or Claude for oil and gas drilling? What petroleum engineers actually need from an AI assistant

If you've typed "ChatGPT for drilling" or "Claude for petroleum engineering" into a search bar, you probably already know the answer is half yes. Ask either one why a shale section keeps going tight and you'll get a decent textbook answer: reactive clays, swelling, mud weight, cavings at the shakers. Ask it about your well, AW-103, yesterday, and it has nothing. It never saw your morning report.

That gap is the whole story. We built Voho Petroleum to close it: a ChatGPT-style assistant for drilling, completions, production and HSE teams that answers from your own records and says which record each answer came from. You can try it at petroleum.voho.ai with three free questions and no account.

What ChatGPT and Claude are genuinely good at on a rig

Credit where it's due. General chatbots are good teachers. A junior engineer can ask how a coiled-tubing acid job is staged, what a gas test interval means on a hot-work permit, or how to read a decline curve, and get a clear explanation in seconds. They're good at rewriting a messy handover note, too, and at turning bullet points into a cleaner procedure draft.

If that's all you need, use them. Nobody needs a specialist tool to learn what overpull is.

Where general chatbots stop

Petroleum work isn't general knowledge. It's specific records: a DDR, a job program at revision B, a reel's fatigue life, a permit that's two hours past its gas test. Here's where a general chatbot runs out:

• It hasn't read your files. You can paste a report into a chat, but next week's chat has forgotten it, and the colleague on the next shift never saw it.
• It can't show its source. An answer about a tight spot is only useful if an engineer can check the record it came from in a few seconds.
• It doesn't hand you a working file. A cost breakdown pasted as text still has to be rebuilt in Excel before anyone approves it.
• Nothing waits for a person. Drafting a work order is easy; making sure nobody posts it to SAP without an engineer's sign-off is the part that matters on a live operation.
• Your data leaves the building. For many Gulf operators, sending well data to a consumer chatbot isn't a policy question that's still open. It's already answered, and the answer is no.

How Voho Petroleum works

It looks like the chat you already know, on purpose. The difference is what sits around the chat box.

Projects: your well file, read on every question

Start a project for a well, a campaign or a job, and add its files: DDRs, job programs, procedures, spreadsheets. Every chat in that project searches them, and every answer lists which files it used. You can also keep documents in Data, your own library, which is searched in every chat.

Answers laid out like a working document

Voho Petroleum leads with the answer, then gives it in full: steps, a table where figures line up, units on every number. When it had to assume something, it says so under "A couple of things worth flagging". When a detail only you know would sharpen it, it asks. It's told never to invent a standard, a source or a vendor price, and to label any planning figure as illustrative.

Excel and Word files, not walls of text

Ask for an NPT breakdown or a cost sheet and you get an actual Excel workbook with formulas, or a Word document for a procedure or a report. The file stays with the chat and opens in the preview.

Approvals: engineers decide

Anything Voho drafts, from a chat answer, a check or an alert, waits in Approvals. A person approves it, edits it or rejects it, and every decision is logged with who and when. In a deployment, an approved action posts to the system it belongs to, like SAP or the permit system. Nothing skips the queue.

Talk to it

Press call and ask out loud in English, Arabic or Urdu. It answers from the same records as the chat, which helps when your hands are busy and the question is quick.

See it on a demo field first

The easiest way to judge it is the Al Waha field, a fictional demo asset with 4 wells, AW-Rig 2 and CT Unit 7, wired to made-up daily drilling reports, real-time drilling data, SAP maintenance, permits to work, an HSE log and a production historian. Because it's fictional, anyone can poke at it.

Questions worth trying there:

• "AW-103 had two tight spots in 48 hours. What do the records say, and who needs to look at it?"
• "Is the coiled tubing reel for the AW-102 job fit to run?" (The reel's at 61% fatigue life and the wall survey is 9 months old, so it should push back.)
• "Is the hot work on the AW-101 flowline safe to continue right now?" (The last gas test was 4 hours ago against a 2-hour interval.)
• "Break down AW-Rig 2's NPT for the last 30 days and say what to fix first." Then ask for it as an Excel sheet.

That's the test we'd run on any AI for drilling: does it catch the thing a tired engineer might miss, and can you check its answer against the record in fifteen seconds?

What it doesn't do (yet)

We'd rather say this now than have you find it.

• Your enterprise systems aren't one click away. WITSML, WellView, OpenWells, AVEVA PI, SAP, Maximo, your permit-to-work system, SharePoint and the rest are connected by Voho's engineers inside your network, one deployment at a time. On a free account, the sources are your project files and your Data library.
• It's an assistant, not a well-control system. It drafts, checks and explains. People make the calls, which is why the approvals queue exists.
• The free tier is small: three questions without an account. Sign up free to keep chatting, start projects, keep documents and get Excel and Word files.

Try it

Open petroleum.voho.ai, ask one of the starter questions, then ask one about a job you're actually working on. If you want it running against your own DDRs, historian and SAP, on enterprise terms with Saudi data residency or on-premise, book a call and we'll scope it. For the wider picture of what makes industrial AI survive past the pilot, read AI in oil and gas: what survives contact with a Gulf operation.