B2B SaaS & software
We asked four AI assistants which software to buy.
They almost never agreed.
Ten buying questions asked by teams choosing software, run through four AI assistants, with every product they named and every source they cited.
Original research · 10 questions × 4 assistants
Data collected · Published
What these numbers mean
Our methodology: we asked 10 buyer-intent questions of ChatGPT, Claude, Gemini and Perplexity, recorded every business named and every website cited in each answer, and repeated the run to measure how much each assistant varies with itself. Nothing here is modelled or estimated. Every figure below is counted from those answers, and the raw dataset is published under CC BY so you can recount them.
- All four used the same website
- When an assistant answers, it reads websites and lists them as its sources. We checked whether all four assistants read at least one website in common. On 0 of the 10 questions, they did. On the rest, there was not a single website all four had looked at. If you see a 0 on a question below, that is what it means, not that they read nothing.
- Used by one assistant only
- We took every website the assistants used across all 10 questions. 85% were used by one assistant and by no other. In other words, most of what they read, they read alone.
- Different websites
- How many separate websites came up in total. Each one counts once, no matter how often it appears. 189 sites for 10 questions means they are pulling from a very wide, scattered set of places rather than a few trusted ones.
You can check all of this yourself. Every question we asked is further down, along with the businesses each assistant named and which of the four used each website. Nothing here is estimated.
Every assistant got exactly the same question. No follow-ups, nothing about who was asking. We wrote down which businesses each one named and which websites it read to get there.
There is no single “AI answer” for your category. Ask two assistants the same question and they agree on about 7% of the websites they read. Ask four and you get four different shortlists, built from four different sets of sources.
That matters if you are trying to get your business recommended. “Are we visible in AI search?” is really four questions, and the answers barely relate to each other. Doing well in Perplexity tells you almost nothing about whether ChatGPT knows you exist.
First, the obvious objection
Ask an AI assistant the same question twice and you can get two different answers. So before saying these four disagree with each other, we had to rule out that each one simply disagrees with itself.
So we asked 10 of the questions again, two from each of the five industries, 3 times on every assistant, and compared what they read each time. Two numbers came out of it:
Same assistant, different runs
51%How much an assistant agrees with itself when asked again
Different assistants, same run
7%How much two assistants agree with each other
So an assistant is 7.5× more likely to agree with itself than with the others. The randomness is real, but it is far too small to explain the gap. They disagree because they are different, not because we got unlucky.
The re-runs turned up something else: some assistants are far steadier than others. Ask Perplexity the same question twice and about 72% of the websites it reads are the same. Ask ChatGPT and only 32% are. Most of what it reads the second time is new.
So checking “what does AI say about us?” once tells you what happened that one time. It is not a measurement, especially on ChatGPT.
The websites they read most
Across all 10 questions. Anything marked directoryis a listing or review site rather than a business's own website.
youtube.com6reddit.com6thedigitalprojectmanager.com4lindy.ai3zapier.com3serviahelpdesk.com3maropost.com3emailtooltester.com3leadsmonky.com3About 10% of what they read was a directory rather than the website of a business they recommended. Take the exact figure loosely. It moves depending on where you draw the line between a directory and a publisher. The pattern is the point: when an assistant needs to work out who is good, it goes to a page listing lots of businesses, not to any one business's own site.
Question by question
Each colour is an assistant. A row with one dot lit is a website only that one assistant read.
Question 1
Best CRM for a small startup?
ChatGPT
HubSpot CRM
- Zoho CRM
Salesforce Essentials
- Pipedrive
- Freshsales
- +2 more
named these, read no websites
Claude
- Zoho CRM
- Pipedrive
- Freshsales
- Capsule
- +2 more
read 8 websites
Gemini
- Less Annoying CRM
- Pipedrive
- Freshsales
- EngageBay
- +5 more
read 9 websites
Perplexity
- Zoho CRM
read 5 websites
Which websites they used, and who used them · top 8 of 19
quickbase.com | 2/4 | |
salesforce.com | 2/4 | |
techradar.com | 2/4 | |
baserow.io | 1/4 | |
blog.salesflare.com | 1/4 | |
capterra.comdirectory | 1/4 | |
fayedigital.com | 1/4 | |
keap.com | 1/4 |
- 19 websites used
- 17 businesses named
- 11% of them were directories
- 0 used by all four
Question 2
Recommend an AI note-taking app for meetings.
ChatGPT
- Otter.ai
- Evernote
Microsoft OneNote
Notion
- Fireflies.ai
named these, read no websites
Claude
- Otter.ai
- Fireflies.io
Avoma
- Granola
- Fellow
named these, read no websites
Gemini
- Fireflies.ai
- Otter.ai
- Bluedot
Jamie
- Granola
- +7 more
read 12 websites
Perplexity
- Otter
Lindy
- Granola
- Fireflies
- +1 more
read 14 websites
Which websites they used, and who used them · top 8 of 22
lindy.ai | 2/4 | |
metaview.ai | 2/4 | |
reddit.comdirectory | 2/4 | |
voicetonotes.ai | 2/4 | |
assemblyai.com | 1/4 | |
avoma.com | 1/4 | |
digitalocean.com | 1/4 | |
krisp.ai | 1/4 |
- 22 websites used
- 19 businesses named
- 9% of them were directories
- 0 used by all four
Question 3
Best help desk software for a SaaS company?
ChatGPT
Zendesk
- Freshdesk
- Intercom
Help Scout
- Jira Service Management
- +2 more
named these, read no websites
Claude
- Freshdesk
- Intercom
- Fin
- +2 more
read 8 websites
Gemini
- Freshdesk
- Zoho Desk
- Intercom
- +4 more
read 9 websites
Perplexity
- Freshdesk
- Intercom
- +1 more
read 7 websites
Which websites they used, and who used them · top 8 of 20
serviahelpdesk.com | 3/4 | |
usepylon.com | 2/4 | |
zendesk.com | 2/4 | |
clearfeed.ai | 1/4 | |
deelo.ai | 1/4 | |
desk365.io | 1/4 | |
dragapp.com | 1/4 | |
ever-help.com | 1/4 |
- 20 websites used
- 14 businesses named
- 5% of them were directories
- 0 used by all four
Question 4
Top email marketing platform for e-commerce?
ChatGPT
Klaviyo
Mailchimp
Drip
Omnisend
- ActiveCampaign
named these, read no websites
Claude
Maropost Marketing Cloud
- Braze
- +1 more
read 7 websites
Gemini
- ActiveCampaign
- +4 more
read 6 websites
Perplexity
- Braze
read 10 websites
Which websites they used, and who used them · top 8 of 16
emailvendorselection.com | 3/4 | |
maropost.com | 3/4 | |
drip.com | 2/4 | |
insiderone.com | 2/4 | |
sequenzy.com | 2/4 | |
brevo.com | 1/4 | |
emailtooltester.com | 1/4 | |
encharge.io | 1/4 |
- 16 websites used
- 10 businesses named
- 13% of them were directories
- 0 used by all four
Question 5
Best project management tool for remote teams?
ChatGPT
- Trello
- Asana
Monday.com
Slack
ClickUp
- +3 more
named these, read no websites
Claude
- Asana
- Trello
- Jira
- Basecamp
- +2 more
read 8 websites
Gemini
Wrike
Taskade
- +4 more
read 32 websites
Perplexity
- Taro
Miro
read 4 websites
Which websites they used, and who used them · top 8 of 41
thedigitalprojectmanager.com | 3/4 | |
worksbuddy.ai | 2/4 | |
activitytimeline.com | 1/4 | |
amoeboids.com | 1/4 | |
appiod.com | 1/4 | |
atlassian.com | 1/4 | |
box.com | 1/4 | |
brainstorminc.com | 1/4 |
- 41 websites used
- 16 businesses named
- 2% of them were directories
- 0 used by all four
Question 6
Recommend a cold email outreach tool.
ChatGPT
- Mailshake
- Lemlist
Hunter Campaigns
- Reply.io
Saleshandy
named these, read no websites
Claude
- Instantly
- Smartlead
- Lemlist
Apollo
- +2 more
read 8 websites
Gemini
- Instantly.ai
- Smartlead.ai
- Lemlist
- Apollo.io
- +4 more
read 6 websites
Perplexity
- Snov.io
- Instantly.ai
- GMass
- Reply.io
read 9 websites
Which websites they used, and who used them · top 8 of 15
leadsmonky.com | 3/4 | |
saleshandy.com | 3/4 | |
snov.io | 3/4 | |
emailtooltester.com | 2/4 | |
mailforge.ai | 2/4 | |
amplemarket.com | 1/4 | |
brandid.app | 1/4 | |
growmeorganic.com | 1/4 |
- 15 websites used
- 15 businesses named
- 13% of them were directories
- 0 used by all four
Question 7
Best customer feedback / survey tool?
ChatGPT
- SurveyMonkey
Typeform
- Google Forms
- Qualtrics
- Zoho Survey
- +3 more
named these, read no websites
Claude
Zonka Feedback
- Qualtrics
- AskNicely
- Refiner
- +5 more
read 7 websites
Gemini
- Qualtrics CustomerXM
- Medallia Experience Cloud
- InMoment
SurveySparrow
- +7 more
read 8 websites
Perplexity
Enterpret
- Qualtrics
- Medallia
- InMoment
SmartSurvey
read 2 websites
Which websites they used, and who used them · top 8 of 15
surveysparrow.com | 2/4 | |
typeform.com | 2/4 | |
averybit.com | 1/4 | |
chattermill.com | 1/4 | |
enterpret.com | 1/4 | |
featureupvote.com | 1/4 | |
formgrid.com | 1/4 | |
getbeamer.com | 1/4 |
- 15 websites used
- 24 businesses named
- 7% of them were directories
- 0 used by all four
Question 8
Top no-code website builder for startups?
ChatGPT
- Wix
- Squarespace
- Webflow
- Shopify
- Weebly
- +1 more
named these, read no websites
Claude
- Squarespace
- Wix
- Hostinger
- Carrd
- Bubble
- +1 more
read 8 websites
Gemini
- Webflow
- Wix
- Squarespace
- Shopify
- Carrd
- +6 more
read 9 websites
Perplexity
- Squarespace
- Shopify
- Webflow
- Framer
- Wix Studio
read 8 websites
Which websites they used, and who used them · top 8 of 24
sitebuilderreport.com | 2/4 | |
anything.com | 1/4 | |
baserow.io | 1/4 | |
blaze.tech | 1/4 | |
callgear.com | 1/4 | |
codeless.co | 1/4 | |
elementor.com | 1/4 | |
emergent.sh | 1/4 |
- 24 websites used
- 14 businesses named
- 13% of them were directories
- 0 used by all four
Question 9
Best product analytics tool for tracking usage?
ChatGPT
- Mixpanel
Amplitude
- Heap
- Google Analytics
Pendo
- +2 more
named these, read no websites
Claude
- Mixpanel
- Heap
- PostHog
- +1 more
read 8 websites
Gemini
- Mixpanel
- Heap
Woopra
- +4 more
read 6 websites
Perplexity
- Mixpanel
- Heap
- Google Analytics 4
read 4 websites
Which websites they used, and who used them · top 8 of 14
amplitude.com | 3/4 | |
cpoclub.com | 2/4 | |
visionlabs.com | 2/4 | |
booboone.com | 1/4 | |
cleverx.com | 1/4 | |
gainsight.com | 1/4 | |
guideflow.com | 1/4 | |
learn.g2.comdirectory | 1/4 |
- 14 websites used
- 12 businesses named
- 14% of them were directories
- 0 used by all four
Question 10
Recommend a social media scheduling tool for agencies.
ChatGPT
- Hootsuite
Buffer
Sprout Social
- Later
- SocialBee
named these, read no websites
Claude
- Sendible
Planable
- Iconosquare
read 8 websites
Gemini
- Hootsuite
- Sendible
Pallyy
- +3 more
read 9 websites
Perplexity
- Sendible
- Agorapulse
- SocialPilot
read 13 websites
Which websites they used, and who used them · top 8 of 22
planable.io | 3/4 | |
sintra.ai | 3/4 | |
sproutsocial.com | 3/4 | |
thecmo.com | 3/4 | |
adamconnell.me | 1/4 | |
authencio.com | 1/4 | |
buffer.com | 1/4 | |
cloudcampaign.com | 1/4 |
- 22 websites used
- 13 businesses named
- 14% of them were directories
- 0 used by all four
What we'd take from this
Treat each assistant separately. They agree on 7% of their sources. That is not four windows onto one ranking. It is four audiences who have read different things. Getting named by one tells you nothing about the other three.
Checking once proves nothing, especially on ChatGPT. When we asked again, only 32% of the websites ChatGPT read were the same. One screenshot of a good answer, or a bad one, is normal variation. If you want to know whether something is really changing, you have to check repeatedly.
The pages that list you may matter more than your own. Assistants read directories far more than they read the websites of the businesses they went on to recommend. Missing from those directories is a different problem from having a weak website, and usually an easier one to fix.
None of this is a ranking. An assistant naming a business tells you what that assistant read, not whether the business is any good. Plenty of excellent firms appear nowhere here. That absence is the finding, not a verdict on them.
Is your site one of the ones being read?
Greater Than Services runs this same test on your website, across every assistant, and shows you which questions you are missing from and what to fix.
Download the raw data (CSV), every question, what each assistant named, and every website it read. Free to reuse under CC BY 4.0, so quote it, chart it, or check our arithmetic. Credit Greater Than Services and link back.
Method.Each question was sent to GPT-4o with web search, Claude Haiku 4.5 with web search, Gemini 2.5 Flash with Google Search grounding, and Perplexity Sonar. One question per request. No follow-ups, and nothing about who was asking or where they were, only the wording of the question itself. The websites come from each assistant's own list of sources; the business names were pulled out of the answer text. We kept every raw answer. 10 of the questions were asked again, 3 times each, to check how much answers move on their own (see above). All of this is a snapshot: ask these questions today and you will not get the same results. Nothing here is a ranking or a recommendation, and naming a business is not an endorsement of it.