1. Home
  2. Research
  3. Who AI assistants name

Who AI assistants name as top AI development companies

AI assistants named LeewayHertz in 16 of 48 answers about hiring an AI firm, more than any other firm. Accenture came next with 14, and no firm was named in more than a third of the answers. On 14 November 2026 we asked ChatGPT, Claude, Gemini and Perplexity the same 12 questions a US buyer asks when hiring for AI work. This page counts the answers. It does not rank anyone, ourselves included.

Published 15 November 2026Checked 14 November 202648 answers, 4 assistantsCadenceWorks research

Answers out of 48 that named LeewayHertz, the firm named most often on 14 November 2026. Accenture came next with 14. No firm was named in more than a third of the answers, and none named CadenceWorks.

What we asked, and how we counted

Here is the method, step by step, so you can repeat it.

  1. We wrote 12 questions the way a US buyer asks when hiring for AI work. They are in the second table below.
  2. We sent each question once to four assistants between 17:36 and 17:48 UTC on 14 November 2026. That made 48 answers. We collected them through DataForSEO, a data supplier.
  3. Every assistant answered from the United States. ChatGPT and Gemini were set to English and the US. Claude and Perplexity searched the web from the US.
  4. We asked ChatGPT all 12 again, minutes after the first run, to see how much its list changes. Those 12 answers are kept out of the 48.
  5. A script pulled out the firms each answer gave as options. For ChatGPT that is the brand list it returns with each answer. For the other three it is a bold label or a heading, which is how those answers set out their picks.
  6. We trimmed each label. We cut it at its first bracket, dash or colon, and kept it when one to four words were left. A bold group of five or more names, such as Claude's “Large global firms (Accenture, Deloitte, IBM, Cognizant, HCLTech)”, did not count.
  7. We counted each firm once per answer. IBM includes IBM Consulting, and Deloitte includes Deloitte Digital.
  8. We ran the tally again on 15 November 2026 and checked it against every firm name in bold text or headings. The script and its output are saved with our research files.

All four assistants cited web pages in their answers. This table shows the model behind each one and how we reached it. An API is the direct line that software uses to reach a model, as opposed to its chat app.

Scroll the table sideways to see every column.

AssistantModelHow we reached it
ChatGPTgpt-6DataForSEO's LLM Scraper, which reads the results page of ChatGPT's own search mode. Web search forced on.
Gemini3.5 Flash-LiteDataForSEO's LLM Scraper for Gemini.
ClaudeClaude Sonnet 5.5The model's API with web search on, through DataForSEO's LLM Responses.
Perplexitysonar-proThe model's API with web search on, through DataForSEO's LLM Responses.

The 12 questions, written the way a buyer asks:

No.Question
1What are the best AI development companies in the US for a mid-size business?
2Which AI consulting firms should a mid-market manufacturer talk to?
3Top AI agent development companies in 2026
4Who builds custom AI agents that work inside an ERP like NetSuite or SAP Business One?
5Best agentic AI development company for enterprise workflows
6Recommend a generative AI development company that can integrate with our existing CRM and ERP
7Which companies offer AI engineering services that build and run the software, not just strategy?
8Best data engineering companies to get our data ready for AI
9Who can build an AI agent on top of our custom-built ERP that has no API?
10AI integration companies for Salesforce and HubSpot
11Which firms help distributors use AI for order entry and quoting?
12Best custom software development companies that also build AI

We asked 16 more questions the same day, and none of them is in the counts on this page. Ten were on cost and process, and 3 asked for a definition. One asked what AI agents do in manufacturing and distribution, one named a rival firm and one named us.

Who was named most

Thirteen firms were named in 6 or more of the 48 answers. The last four columns show how many of each assistant's 12 answers named the firm.

Scroll the table sideways to see every column.

FirmAnswers (of 48)ChatGPTClaudeGeminiPerplexity
LeewayHertz163364
Accenture144424
Simform103232
EPAM Systems105023
Azumo103223
IBM93303
Thoughtworks83113
HatchWorks AI82114
Deloitte80323
Markovate70322
Slalom63111
ScienceSoft61122
Cognizant63201

A firm counted only when an answer offered it as an option, as steps 5 to 7 above set out. A passing mention inside a sentence did not count.

Count every mention anywhere in the text and the numbers rise. Accenture and LeewayHertz reach 20 answers each. Some answers mention a firm in passing, and one ChatGPT answer named Accenture as a firm it would not start with.

The question changes the list

The firms at the top of the table did not appear on every kind of question. Here is what each group of questions drew.

  • AI development for a mid-size business (questions 1 and 12). All four assistants named HatchWorks AI on question 1. LeewayHertz and ScienceSoft were named by all four on question 12.
  • AI agents for large companies (3 and 5). Large integrators led. Accenture was named by all four on question 3 and by three on question 5. IBM was named by three on both.
  • Firms that build and run the software (7). Tribe AI and HCLTech were the only firms named by three of the four.
  • Getting data ready for AI (8). Accenture and Tiger Analytics were named by all four.
  • A mid-market manufacturer looking for consultants (2). No firm was named by more than two. Kalypso and West Monroe were each named twice.
  • Generative AI joined to a CRM and ERP (6). No firm was named by more than one assistant.
  • Agents inside NetSuite or SAP Business One (4). None of the 13 firms in the table was named as an option. Perplexity mentioned LeewayHertz and Markovate only inside a group of nine names, which our rule does not count. The answers named NetSuite and SAP partners, firms that build ERP agents, and order-entry software. Gemini led with the vendors' own tools, such as Oracle's AI Connector Service and SAP's Joule Studio.
  • A custom ERP with no API (9). Claude and Gemini named no service firm. Both described kinds of builder instead, and both named screen automation, where software works the ERP's screens the way a person does. ChatGPT and Perplexity named five firms each, with none in common.
  • Salesforce and HubSpot (10). ChatGPT and Claude named partners of those platforms, and Coastal Cloud appeared in both. Perplexity named HubSpot agencies such as SmartBug Media, plus tools such as Workato. Gemini named no firm, only the platforms' own AI, Agentforce and Breeze.
  • Order entry for distributors (11). Software products, not service firms. Canals, Proton.ai and Conexiom were named by all four.

Ask twice, get a new list

We asked ChatGPT all 12 questions a second time, minutes after the first. On average the two answers shared a third of their names. Overlap here is the names in both answers divided by all the names in either: 1.00 means the same list, 0 means nothing in common.

Scroll the table sideways to see every column.

QuestionNames, run 1Names, run 2In bothOverlap
16640.50
25620.22
3101480.50
4111280.53
571130.20
64310.17
7191680.30
87850.50
95510.11
108520.18
111312100.67
1210830.20

The average across the 12 questions was 0.34. The custom ERP question kept 1 name of 9. The distributor question, where the answers named software products, kept 10 of 15.

What the assistants read

DataForSEO returns the searches an assistant ran for ChatGPT and Claude only. Across both ChatGPT runs, ChatGPT ran 51 web searches for its 24 hiring answers. Claude ran 23 for its 12.

  • 26 of those 74 searches held the word “best” or “top”. Claude's first search on question 1 was “best AI development companies USA 2026”.
  • None was limited to one website. It was different on 9 October 2026. That day we asked ChatGPT 18 ERP and CRM questions for five countries, through its API. It limited 72 of its 108 searches to a single site, mostly the ERP and CRM vendors' own sites. The questions and the route differ, so the two runs are not a like-for-like test. It is why we now read ChatGPT through its own search mode.
  • 47 of the 48 answers cited at least one web page. The sites cited most were uvik.net (7 answers) and reddit.com (5). uvik.net has blog posts that list custom AI development companies. One Reddit thread's address reads “top 10 AI development companies in USA 2026”. rtslabs.com was also cited in 5 answers. It is the site of RTS Labs, an AI firm, and one of the cited pages lists top AI agent development companies.

RTS Labs was named in 4 answers, 3 of them Gemini's.

What it means if you are choosing a firm

Our reading of the counts above:

  1. A name in an answer reflects the pages the assistant read that day. The pages cited most were lists published by firms and forum threads. It is not a test of anyone's work, and we did not assess any firm.
  2. Ask more than once, and ask more than one assistant. ChatGPT kept about a third of its names between two runs a few minutes apart.
  3. Name your system in the question. On the NetSuite and SAP Business One question, none of the 13 most-named firms made an assistant's list of options. On the Salesforce and HubSpot question, only Claude named any of them.
  4. Open the pages the assistant cites and check who wrote them before you take its shortlist.

What we did not measure

  • Microsoft Copilot, DeepSeek, Muse and Grok. DataForSEO's answer endpoints cover ChatGPT, Claude, Gemini and Perplexity, and our requests for the others failed on 14 November 2026.
  • Claude and Perplexity as their apps show them. We asked both through their APIs. An API and an app can answer differently, as the October searches above show for ChatGPT.
  • Change over time. Each question went to each assistant once, twice for ChatGPT, on one day. We have not repeated it yet.
  • Other countries and languages. English questions, asked from the United States only. We also collected Google AI Mode answers that day, and they are not in these counts.

Questions about this count

Which 10 AI development companies did the assistants name most?

The 10 firms named most in our 48 answers of 14 November 2026 were LeewayHertz (16 answers), Accenture (14), Simform, EPAM Systems and Azumo (10 each), IBM (9), Thoughtworks, HatchWorks AI and Deloitte (8 each), and Markovate (7). No firm was named in more than a third of the answers. This counts which firms the assistants found in their sources that day. It is not a test of their work.

Can you give me a list of AI development companies that assistants name?

Yes. Thirteen firms were each named in 6 or more of our 48 answers: LeewayHertz, Accenture, Simform, EPAM Systems, Azumo, IBM, Thoughtworks, HatchWorks AI, Deloitte, Markovate, Slalom, ScienceSoft and Cognizant. The table on this page shows how often each assistant named each one. On the question about agents inside NetSuite or SAP Business One, none of these 13 made an assistant's list of options.

What are the top AI consulting companies, according to AI assistants?

Accenture, IBM and Deloitte were the large consulting firms named most across all 12 questions. Accenture was in 14 of 48 answers, IBM with IBM Consulting in 9 and Deloitte in 8, and Claude grouped all three under “Large global firms”. The assistants did not agree on the consulting question itself. For “Which AI consulting firms should a mid-market manufacturer talk to?”, no firm was named by more than two of the four. Kalypso and West Monroe were each named twice.

Who builds AI agents inside NetSuite or SAP Business One, according to assistants?

The four answers named NetSuite and SAP partners, firms that build ERP agents, and order-entry software. None of the 13 most-named firms made an assistant's list of options on that question. Perplexity mentioned two of them, LeewayHertz and Markovate, only inside a group of nine names. Gemini started with the vendors' own tools, such as Oracle's AI Connector Service and SAP's Joule Studio. So name your system when you ask an assistant for firms.

Which websites do AI assistants cite for top AI development companies?

The three sites cited most were uvik.net, in 7 answers, and reddit.com and rtslabs.com, in 5 each. Of our 48 answers, 47 cited at least one web page. uvik.net has blog posts that list custom AI development companies. rtslabs.com is the site of RTS Labs, an AI firm. Open the cited pages and check who wrote them.

Does ChatGPT name the same firms if you ask twice?

Only about a third of them. We asked the ChatGPT app all 12 questions twice, minutes apart. The average on our overlap measure was 0.34, where 1.00 means the same list. The lowest was 0.11, for a custom ERP with no API (no way for other software to connect). The highest was 0.67, for order entry at distributors.

Why do ChatGPT, Claude, Gemini and Perplexity name different firms?

Each one searches the web and builds its answer from the pages it finds, and those pages differ. LeewayHertz was named in 6 of Gemini's 12 answers and 3 of ChatGPT's. EPAM Systems was named in 5 of ChatGPT's and none of Claude's.

How did you decide that an answer named a firm?

A firm counted once per answer when the answer put it in its list of options. For ChatGPT that is the brand list it returns with each answer. For the other three it is a bold label or heading of one to four words. A passing mention inside a sentence did not count, and neither did a bold group of five or more names. The script and its output are saved with our research files.

How should I check a shortlist an AI assistant gives me?

Ask more than once, and ask more than one assistant. ChatGPT kept about a third of its names between two runs a few minutes apart. Put your system's name in the question. On the NetSuite and SAP Business One question, none of the 13 most-named firms made an assistant's list. Then open the pages the assistant cites and check who wrote them before you trust its list.

Is CadenceWorks in these counts?

No. None of the 60 hiring answers, including the second ChatGPT run, mentioned CadenceWorks. We checked the answer text, ChatGPT's brand lists and the cited sites for our name. This note does not rank us, and we did not rate any firm. It counts what four assistants said on one day.

Which AI assistants did you not test?

Microsoft Copilot, DeepSeek, Muse and Grok. Our data supplier, DataForSEO, had no way to collect their answers on 14 November 2026. We asked Claude and Perplexity through their APIs, the direct line that software uses, so their answers may differ from what their own apps show.

Sources

The answers are our own data, collected on 14 November 2026 and tallied again on 15 November with a script saved beside the output. DataForSEO's documentation was read on 15 November 2026.

Choosing a firm for AI work?

Tell us the job and the systems it has to work with. A person on our team replies by email.

Tell us the job