The AI Software Index 2026
What ChatGPT and Gemini actually recommend when a buyer asks for software. Measured, not guessed.
What we did
- ·We wrote ~80 questions per software category, the way real buyers ask them. Things like "What's the best CRM for a small business?" or "I'm a freelancer and need invoicing software."
- ·We asked them on the real ChatGPT and Gemini apps: the same thing a buyer sees, not a developer API.
- ·For every answer we recorded which brands were named, which one was recommended first, and which websites the AI cited as sources.
What we found
Seven things the data says, in plain English.
- 1ChatGPT and Gemini disagree a lot. In 1 out of 3 categories, they name a different #1 product. Which AI your buyer uses changes who gets recommended.
- 2ChatGPT links to the winner's own website. Gemini doesn't. When ChatGPT recommends a product, it cites that product's own site 53% of the time. Gemini: only 13%. Gemini relies on third-party pages instead.
- 3Gemini trusts small blogs, not big media. Its most-cited "independent" sources are blog posts written by other software companies, often about categories they have nothing to do with.
- 4Review sites barely matter. G2, Capterra, and TrustRadius together get only 5% of all citations.
- 5The answer changes when you rephrase the question. In some categories one brand always wins. In others, asking the same thing in different words gives you a different "best" tool almost every time.
- 6Saying who you are changes the answer. "I'm a freelancer" and "we're an enterprise" get different recommendations. In some categories a brand jumps from winning 34% of answers to 90%.
- 7US and UK buyers get different winners. The #1 recommendation flips between US- and UK-phrased questions in about half of the categories we could measure.
The most AI-visible software brands
How many category leaderboards each brand appears on, and how many categories it actually wins.
Being everywhere and winning are different things. Salesforce shows up on 13 leaderboards and wins none of them. QuickBooks shows up on 7 and wins 5.
Locked up vs. up for grabs
In some categories one brand wins no matter how you ask. In others, the winner changes with the wording.
One clear winner
The websites AI trusts most
The domains cited most often across all ~10,000 answers.
- 1.
techradar.com1,114
- 2.
reddit.com807
- 3.
zapier.com418
- 4.
learn.g2.com342
- 5.
hubspot.com326
- 6.
microsoft.com283
- 7.
zoho.com262
- 8.
gartner.com238
- 9.
atlassian.com206
- 10.
clickup.com176
- 11.
monday.com175
- 12.
guideflow.com158
- 13.
pcmag.com143
- 14.
asana.com141
- 15.
zendesk.com132
- 16.
quickbooks.intuit.com123
All 126 categories
Grouped by what the software is for. Every page has the full AI leaderboard for that category, filterable by engine, buyer type, and region.
Sales and CRM
Where AI sends buyers for pipeline, outreach and closing the deal.
Marketing
Email, automation, SEO, social and the rest of the growth stack.
Customer support and feedback
Help desks, chat, knowledge bases and the tools that collect what customers think.
E-commerce and retail
Storefronts, point of sale, billing and getting orders out the door.
Projects and operations
Planning, task tracking, scheduling and work in the field.
HR and people
Hiring, payroll, benefits and everything after the offer letter.
Finance and accounting
Books, invoices, expenses, tax and getting paid.
IT and security
Passwords, endpoints, access, backups and keeping the lights on.
Collaboration and productivity
Meetings, chat, files, notes and booking time with people.
Developer and automation
Shipping code, tracking bugs and wiring tools together without code.
Websites and analytics
Building the site, measuring who uses it and asking them questions.
Industry software
Software built for one kind of business: clinics, gyms, hotels, schools, churches and more.
How this index is built
The AI Software Index measures what two assistants tell a buyer who asks for software: ChatGPT (gpt-5.5) and Gemini, both read from their live consumer web apps rather than a developer API, so the answers are the ones a buyer actually sees.
For each of the 126 categories we wrote about 80 buyer questions, phrased the way people ask them. Some are generic (“what’s the best CRM?”), some say who is asking (a freelancer, a startup, an enterprise, a nonprofit, an agency, a team on a budget), and some are phrased for US or UK buyers. Each question was answered once, giving 9,978 answers: 5,023 from ChatGPT and 4,955 from Gemini.
From every answer we record which brands were named, which was recommended first (the “top pick”), and which websites were cited. Strings the extractor picked up that are not products, such as follow-up questions and feature names, are removed, and duplicate names for one product are merged.
The study runs quarterly. The current data was measured on 6 August 2026, and every category page shows its own date and question count. It is one snapshot: rephrase a question and the winner can change, which is why the pages report shares across many questions rather than one answer.
Frequently asked questions
AI is already recommending someone in your category.
GetIntel tracks whether it's you, daily on ChatGPT, Perplexity, Gemini and Google AI Overviews, then ships the fixes through your own coding agent.