Bildad OyugiA buyer can now ask ChatGPT which tool to pick and get an answer with sources attached. If your page isn’t one of those sources, you aren’t in that answer.
So I tested what it cites. The short version: ChatGPT and Google draw on different pages, and the question type decides which kind.
What I tested
I asked eight questions a B2B software buyer might ask, with ChatGPT’s web search switched on. Then I recorded every page it cited and compared them with Google’s first page for the closest search.
How the test ran1 Oct 2026
ChatGPT (gpt-4o) with web search, through OpenAI’s API. Each prompt began “Search the web:” so the model would search rather than answer from memory. Google results are US, first page, run the same day.4
Finding 1: ChatGPT’s sources are not Google’s page one
Only 2 of the 57 cited pages also appeared on Google’s first page. One was Intercom’s pricing page. The other was a Userpilot guide on reducing support tickets.
| Question | Pages cited | Also on Google p1 |
|---|---|---|
| Best help desk software for a small B2B SaaS company | 5 | 0 |
| Zendesk vs Help Scout for a 5-person team | 8 | 0 |
| How much does Intercom cost in 2026 | 4 | 1 |
| What is a good first contact resolution rate | 9 | 0 |
| What is generative engine optimization | 9 | 0 |
| Best CRM tools for a 10-person B2B startup | 11 | 0 |
| Best alternatives to HubSpot Marketing Hub | 6 | 0 |
| How to reduce support ticket volume | 5 | 1 |
Take the help desk question. ChatGPT cited Software Advice, Capterra, G2, Ferndesk and Plain. Google’s first page showed Salesforce, Pylon, BoldDesk, Reddit, ClearFeed, Zoho, Kustomer and Zendesk. No page appeared in both.
OpenAI’s help centre explains part of this. ChatGPT search rewrites a question into one or more targeted queries and sends them to its search partners.1 The pages it sees depend on that rewritten query, not on where you rank for yours.
Finding 2: the question decides the source
Each type of question pulled a different type of page.
| Question type | What ChatGPT cited |
|---|---|
| “Best X for Y” | Software directories and publisher listicles |
| “X vs Y” | Small comparison sites, plus the vendor’s own comparison page |
| “How much does X cost” | The vendor’s own pages: Intercom supplied 11 of the 13 citations |
| “What is X” and benchmarks | Glossaries, research papers and Wikipedia |
| “How do I…” | Guides on software companies’ blogs |
Across all 57 pages, the mix looked like this.
| Source type | Pages | Share |
|---|---|---|
| Company blogs, guides and glossaries | 23 | 40% |
| Comparison and review publishers | 18 | 32% |
| Software directories (G2, Capterra, Software Advice, Gartner) | 7 | 12% |
| Reference and research (Wikipedia, papers, research firms) | 5 | 9% |
| The vendor’s own pages | 3 | 5% |
| Forums (Reddit) | 1 | 2% |
Finding 3: small sites get cited
A first-page Google ranking wasn’t the gate. The GEO definition drew on small glossary pages, two research papers and one consultant’s personal site.
Most of the sites cited for Zendesk vs Help Scout are ones few buyers would recognise. What they had was a page built for that exact question.
How to get cited by ChatGPT: 8 steps
- Let OAI-SearchBot crawl your site. OpenAI says sites that opt out of OAI-SearchBot are not shown in ChatGPT search answers. Robots.txt changes take about 24 hours to apply.2 GPTBot is separate: blocking it opts you out of model training, not search.
- Put your content in the HTML. OpenAI’s crawlers don’t run JavaScript, so text that loads after the page renders may never be seen.3
- Match the rewritten query. In my test, “what is the best help desk software for a small B2B SaaS company in 2026?” became “best help desk software for small B2B SaaS company 2026”. Put the audience in your title, and the year when the page is current.
- Build the page the question calls for. Use the table above. A “best X” question needs a listicle, and a “vs” question needs a comparison page.
- Publish your own facts on your own site. ChatGPT answered the pricing question from Intercom’s own pages. Keep a pricing page with current numbers in the HTML.
- Get listed where ChatGPT looks. Directories and comparison sites supplied 25 of the 57 citations. Keep your G2, Capterra and Software Advice profiles current, and pitch the comparison sites in your category.
- Write passages worth quoting. Lead with the answer, then give specific numbers with sources, tables and a visible updated date. In three of the eight answers, ChatGPT named the month its source was published.
- Measure it. Run a fixed list of buyer prompts every month and record whether you are cited and whether you are named. In GA4, visits from ChatGPT show chatgpt.com as the source. In my test, every cited link also carried utm_source=openai.
What this test can’t tell you
Eight questions, one model and one day make a small sample. ChatGPT’s answers change between runs, and the API isn’t identical to the ChatGPT app.
My Google searches were the closest match I could write, not ChatGPT’s exact queries. Treat the patterns as a starting point, then run the same test on your own buyer questions.
For the Google side of the story, read how Helply got cited in 472 Google AI Overviews.
Want to know where you stand?
I run your buyers’ questions through ChatGPT, Perplexity, Gemini and Google’s AI Overviews, then fix what keeps you out of the answer.
See AEO & GEO services