Three AI engines, the same five buying questions. They agreed on the winner and on almost nothing else.

/ 6 min read / By Faz

Most AI search advice treats “AI” as one thing. Get cited, show up in the answer, done. A buyer does not use “AI”. They use one engine, and the one they use decides which pages shaped what they read.

So in November I asked three engines the same five questions, word for word: Google’s AI Mode, ChatGPT and Perplexity. The questions were the kind a B2B buyer asks at the shortlist stage. The best help desk for a 50-person B2B SaaS support team. The best CRM for a 20-person B2B sales team. HubSpot or Salesforce for a Series A startup. The best SOC 2 compliance automation tool for a 60-person startup. The best product analytics tool for a B2B SaaS company. For each answer I recorded which product it put first and which websites it showed.

They mostly agreed on the winner

Question Google AI Mode ChatGPT Perplexity
Help desk, 50-person B2B SaaS team Zendesk Intercom Zendesk
CRM, 20-person B2B sales team HubSpot HubSpot HubSpot
HubSpot or Salesforce, Series A HubSpot HubSpot HubSpot
SOC 2 automation, 60-person startup Vanta Vanta Vanta
Product analytics, B2B SaaS PostHog PostHog Mixpanel

On two questions Google named a small group rather than a single winner, so the table shows the first product it named. On three of the five questions all three engines led with the same product. On the other two, two of the three agreed. The products in each answer’s longer shortlist overlapped heavily too.

If you are the category leader, that is reassuring. A buyer will hear roughly the same verdict whichever engine they open.

They almost never agreed on the sources

Across the fifteen answers, the engines showed 111 different websites. Only 12 of them appeared on more than one engine for the same question, about one in nine. Only 5 appeared on all three.

Every one of those 12 was a vendor’s own site: the help desks, the compliance tools, the analytics products being recommended. Not one third-party page, no review site, no comparison, no list, no Reddit thread, was shown by more than one engine for the same question.

So the verdicts converged, but they were built on almost entirely different evidence. The only pages that reached every engine were the ones the vendors own.

Each engine has its own habit

Google’s AI Mode cast the widest net. It showed 56 websites across the five answers, the most of the three, and most of them were third-party: agencies’ comparison posts, small software review sites, YouTube and Reddit. On the CRM question it showed no vendor sites at all. On HubSpot or Salesforce it showed 20 sources, mostly agencies, consultancies and small software companies writing comparisons. If you want to know who shapes Google’s version of your category, look at who writes comparisons in it. I covered the same pattern in Google’s AI Overviews.

ChatGPT showed the vendors’ own sites. It showed 23 websites across the five, and almost all were the recommended tools’ own websites. On HubSpot or Salesforce it showed no links at all and answered from what it already knew. Its links may be attached to product names rather than being everything it read, so treat them as what the buyer was pointed to, not a full source list. Even so, the pattern is clear: in ChatGPT, the buyer is sent straight to the vendors.

Perplexity read “best tools” lists. It showed 49 websites, and most of its sources were list posts and head-to-head comparisons: “best help desk software”, “best CRM for B2B sales”, “HubSpot vs Salesforce for startups”. Many of those lists were written by software companies in the same category. In two of the five answers, the first source was a “best tools” list written by the product Perplexity then put first. On SOC 2 it was Vanta’s own list of compliance software. On product analytics it was Mixpanel’s own list of analytics tools, and Mixpanel was the one answer where Perplexity disagreed with both other engines.

What this means for you

Your own site is the one source every engine reads. If the only pages that reached all three engines were the vendors’ own, then your product pages, pricing page and core facts are the only material you can be sure each engine sees. Keep them specific, current and easy to quote. That is the job in writing content engines will cite, and it pays on every engine at once.

The third-party work is per engine. The outside pages that shaped each answer were different on each engine. A correction on a Google comparison post will not reach Perplexity, and a place on a list Perplexity reads will not reach Google. Map the sources separately for each engine your buyers use, and decide which ones to earn a place on.

On Perplexity, lists decide a lot, including your own. Vanta and Mixpanel had each published a “best tools” list in their own category, and both lists were the first source behind an answer that led with them. That is not a trick if the list is honest. A fair list that names real competitors and says where each one is better is useful to a buyer and to an engine. A list that ranks you first in every category will read as marketing. The listicle post goes into how to tell which lists are worth anything.

Check which engine your buyers actually use. The engine question matters more than it looks, because the verdict may match across engines while the sources behind it do not. If your buyers mostly use one engine, the sources that matter are that engine’s.

What I got wrong

Two of my older posts said the engines share only about one source in ten. One of them credited “independent studies”. I had not checked the figure myself. My own count, about one in nine, came out close. What I did not know was what the overlap consists of: entirely the vendors’ own sites. The third-party pages did not overlap at all.

I also expected ChatGPT to lean on encyclopedic and third-party references, which is what I had written before. In this test it pointed the buyer at the vendors’ own sites. And I expected Perplexity to lean on community threads. It leaned on lists, and the Reddit threads showed up in Google’s answers instead.

These were five questions, run once each on each engine, from outside the US. ChatGPT and Perplexity were run in a signed-in account, which can affect answers. Treat this as a pattern worth checking in your own category, not a measurement.

Where this leaves you

Whichever engine your buyer opens, they will probably hear a similar verdict. What they read on the way there will be different, and on Google and Perplexity it was built mostly from pages you do not control. Your own site is the one source every engine shares, so make it the best source on your own product. Then do the outside work engine by engine, starting with the one your buyers use.

If you want to see what each engine tells your buyers, and which pages it builds that from, that is part of the engagement, and it runs on a documented method.

Want this run for your B2B SaaS?

Founding pricing for the first 5 clients. Methodology fully public. Month-to-month, cancel anytime.

Apply to work together

1 response

Leave a comment

Your email address will not be published. Required fields are marked *