Use ChatGPT Deep Research to Build a Source-Checked Vendor Shortlist

Evidence label: documentation-based guide with an illustrative example. The sample request and output below are fictional. No ChatGPT Deep Research run is claimed.

A vendor shortlist is useful when you can see why each company is on it. A polished paragraph with no source trail is harder to review than a plain table.

ChatGPT Deep Research is designed for multi-step questions. It can use the public web, uploaded files, selected sites, and eligible connected apps. It creates a research plan that you can review before the task starts, then returns a report with citations or source links.

This guide uses a fictional software purchase. The result is a shortlist for a manager to review, not an automatic recommendation.

Before you start

You need access to ChatGPT Deep Research through the tools menu, sidebar, or /Deepresearch. Availability and usage limits depend on your plan, region, workspace settings, and available app permissions.

Prepare:

A clear buying question.
Two or three source sites you trust.
A short list of criteria.
A decision owner who will review the evidence.

For this example, the reader is comparing three fictional customer-support tools for a 12-person team. The source sites are the vendors’ public documentation and pricing pages. The fictional budget ceiling is 1,500 USD per month. That number is a scenario input, not a market claim.

1. Start with a bounded research question

Open ChatGPT and select Deep research from the tools menu. You can also type /Deepresearch in the prompt box.

Use a prompt that tells the task what to collect, what to ignore, and what the final decision still requires from a person:

Research three customer-support tools for a fictional 12-person team:

- HelpHarbor: https://help.harbor.example/docs
- QueuePilot: https://queuepilot.example/docs
- ReplyDesk: https://replydesk.example/docs

Question: Which option deserves a manager review for a team that needs shared inboxes, role-based access, exportable reports, and a documented API?

Use only the public vendor documentation and pricing pages on those sites. If a requested fact is missing, write "Not found in the reviewed sources". Do not infer a feature from marketing language. Do not recommend a winner.

Return:
1. A comparison table with one row per tool.
2. A source link beside every factual claim.
3. A short evidence note for each criterion.
4. A list of unknowns and questions for the vendors.
5. A final shortlist containing at most two tools, with the reason each remains under review.

Treat the 1,500 USD monthly budget as a scenario constraint, not as a claim about market pricing. Show whether each tool's published pricing can be compared with that constraint. Stop before contacting a vendor, changing a subscription, or making a purchase.

The URLs above are fictional. They illustrate the input shape only. Replace them with real sources you are allowed to review.

2. Review the research plan before it runs

ChatGPT Deep Research proposes a plan before the research begins. Read it before starting the task.

Check that the plan:

Covers all three named vendors.
Uses the source sites you specified.
Separates documented features from missing information.
Includes the comparison table and source links.
Keeps the manager’s decision as a human step.

If the plan starts searching broad review sites, or drops the missing-information rule, edit the plan or stop and rewrite the request. A broader search can be useful for discovery, but it should not silently replace your chosen evidence standard.

3. Review the report as a shortlist, not a verdict

The completed report should give you a structured result with citations or source links. For the fictional example, the desired shape is:

Tool	Shared inboxes	Role-based access	Reports	API	Evidence status
HelpHarbor	Documented	Not found in the reviewed sources	Documented	Documented	Keep under review
QueuePilot	Documented	Documented	Not found in the reviewed sources	Documented	Keep under review
ReplyDesk	Not found in the reviewed sources	Documented	Documented	Not found in the reviewed sources	Hold

Illustrative output, not a generated result. The statuses above are editor-written fictional values. They show how a missing source should remain visible instead of becoming a guessed “No”.

Open each citation that affects the shortlist. Confirm that the linked page says what the table claims, and check whether a plan, region, role, or workspace condition limits the feature. Move disputed or unclear rows into the unknowns section.

Acceptance checks

Keep the shortlist only if all of these checks pass:

Every tool has one row in the comparison table.
Every factual cell has a source link or says that the fact was not found.
A missing fact is not rewritten as a negative feature claim.
The budget is labeled as a scenario constraint.
The report names the criteria that kept each shortlisted tool under review.
The manager still has to choose, request clarification, or reject the shortlist.
No vendor was contacted and no subscription was changed by the workflow.

If the result looks wrong

The report uses unsupported review pages. Edit the plan so it uses the permitted source sites, then run a smaller research task. Keep the first report as a draft, not as evidence.

A feature is marked as available without a clear citation. Change that cell to “Not found in the reviewed sources” and add the question to the vendor follow-up list. Do not rely on the tool’s wording alone.

The task cannot access a connected source. Check whether the app supports Deep Research and whether the account, plan, region, workspace role, and provider permissions allow access. Use an uploaded document or public source as a manual fallback when approved.

Limits and privacy

Deep Research usage varies by plan, and connected apps do not override provider permissions or workspace restrictions. The official documentation says Deep Research uses available read actions from connected apps, not app write actions as part of research. Still, use approved files and sources only. Remove confidential customer or employee data unless your organization’s policy permits the upload and the selected account is approved for it.

Sources

The source-linked competitor brief guide covers a smaller ChatGPT research workflow. Use it when a full Deep Research task is unnecessary.

For more practical AI workflows, join the weekly update.