How Does ChatGPT Deep Research Choose Which Companies to Recommend?

How does deep research work?

A normal ChatGPT answer searches briefly, if at all, as covered in how ChatGPT decides when to search. Deep research is the mode buyers use when a decision matters, such as building a vendor shortlist.

What decides which companies it includes?

OpenAI has not published a ranking method, so this is based on how the product works:

• The buyer's criteria. A brief asking for “ISO 27001-certified payroll providers for 50-person companies in Ireland” leads deep research to look for evidence on each point. A company whose certification, customer size and country coverage are not stated anywhere readable is likely to be dropped or marked as unknown.
• The sources it reaches. Comparison articles, review platforms, directories, analyst reports, news and vendor sites all feed the report. Companies that appear across several of them are easier to include with confidence.
• Specific, checkable facts. Prices or price ranges, integrations, locations, certifications and named case studies can be cited. Vague claims cannot.
• Consistency. When your site, review profiles and directories agree, the report can state facts plainly; conflicts invite caveats or exclusion.
• Recency. Current pages and recent coverage help for fast-changing categories.
• Restricted sources. If the user limits the research to certain sites, anything outside them is ignored.

Why AI answers often cite directories and marketplaces rather than your own site is covered in why AI cites directories instead of your website.

Claude Research versus ChatGPT Deep Research for business recommendations

For a business being evaluated, the two reward the same things: facts that can be found and checked, and presence in the sources buyers trust. The differences matter for access. Both companies' search crawlers and user fetchers need to reach your pages, as covered in the difference between ClaudeBot, Claude-User and Claude-SearchBot, and Claude's web search results have matched Brave's, so a page missing from Brave's index may be harder for Claude to find.

How do you become easy to recommend?

1. Publish the facts buyers screen on: pricing or price ranges, service areas, team size, integrations, certifications, turnaround times and contract terms, on crawlable pages.
2. Back claims with evidence: case studies with named industries and numbers, security documentation, and customer logos you have permission to show.
3. Keep third-party profiles complete and consistent on the review platforms, directories and marketplaces in your category.
4. Publish fair comparison pages against the alternatives buyers already know.
5. Allow the crawlers that fetch pages for deep research and Claude Research.
6. Earn coverage in the analyst notes, industry publications and “best of” articles that researchers find first.

Which pages on your site does it need?

A vendor brief usually asks the same questions, so the same pages do most of the work:

• Pricing, with ranges and what each price includes, even if final quotes vary.
• Services or features, one page per main offering, in plain terms.
• Locations and coverage, stating the cities, regions or countries you serve.
• Proof: case studies, client types and results.
• Security and compliance: certifications, data location and policies, for B2B buyers.
• Integrations or partners, named individually.

Keep these as normal HTML pages linked from your navigation. Facts locked in PDFs, behind forms or loaded by scripts are harder for any research agent to read and cite.

How can you test it?

Run deep research yourself with a realistic buyer brief, such as “Shortlist five managed IT providers for a 40-person law firm in Manchester, with pricing, security certifications and client examples.” Read the report and its sources: which companies were included, which sources supported each claim, and what the report said was unknown. Repeat with Claude Research and with sources restricted to your category's main review site. Competitors and buyers may already be running similar briefs to audit you. How to run simpler checks is covered in how to check whether ChatGPT recommends your business.

We have not run a systematic study of deep research recommendations. This page is based on OpenAI's and Anthropic's product documentation and announcements.

What might this look like in practice?

Consider an illustrative bookkeeping firm for restaurants, Tallow & Ledger, that was never included when buyers used deep research to shortlist providers. Running a brief itself, the owner found the report listed three competitors whose websites stated prices, software used and cities served, while Tallow & Ledger's site said only “tailored packages”. Its directory profile also listed an old address. The firm published price ranges, a page on the restaurant software it supports, and two case studies with results, and corrected its profiles. A month later, the same brief included Tallow & Ledger, citing its pricing page and a review profile.

What do people ask about deep research and business recommendations?

Can I pay to be included in deep research reports?

No. Reports are built from the sources the research finds; ads are separate from ChatGPT's answers.

Does deep research read my whole website?

It reads the pages its searches lead to. Clear, linked pages for pricing, services and proof are the ones most likely to be read.

Why was a competitor included and not us?

Usually because their facts matched the brief and could be verified. Compare what the report cited for them with what exists for you.

Can buyers limit deep research to certain sites?

Yes. Since February 2026 they can restrict web searches to trusted sites and connect apps.