Who Do AI Engines Recommend for Conversational AI Consulting?
In this article
Most advice about AI citations is opinion. Nobody publishes the score. So we built a scoreboard and ran it. This is the first edition. That includes the parts where ICX loses.
This is the first ICX AI Citation Index. We collected it on September 12, 2026. It answers one question that a buyer would ask. Someone asks an AI engine about conversational AI consulting. Who gets named?
What is the ICX AI Citation Index?
The index is a fixed set of 28 prompts. ICX runs them through live AI engines on a schedule. Then a script scores each answer on one thing. Does it name or cite a given brand?
The 28 prompts split into two groups. Twenty-two are category prompts. A category prompt is an open question with no brand. Here is an example. “Who are the best conversational AI consultants for enterprise customer service?” It names no firm. These prompts show whether a buyer can discover a firm. Six are branded prompts that name ICX or its founder. They test one thing. Does an engine that knows the name describe the firm correctly?
We wrote the prompt set in June 2026. It was a tracking list. The set has not changed since then. It covers the buyer identity question and SaaS support cost. It covers service-specific hiring and ROI. It covers Dallas-Fort Worth local search and topical authority. The full list is in the data file below.
We explained the ideas before. See what AEO and GEO actually mean. Also see how Google AI Overviews pick their sources. Now we put a number on our own position.
How was the September 2026 index measured?
Two engines answered every prompt on September 12, 2026.
The first engine was Perplexity Sonar. It runs a web search for every answer. It returns sources. The second engine was OpenAI’s GPT-4o-mini. The OpenRouter web search plugin was on. With the plugin, it searched the live web first. We called both engines through the OpenRouter API. Both got the same wording. Answers had a 700-token cap. Each prompt ended with one extra sentence. It asked the engine to include its URLs.
ICX scored every answer in two ways. Named means the text has one of three names. These are Intelligent CX Consulting, ICX, or Christi Akinwumi. Cited means the sources list a URL from one of two sites. These are intelligentcxconsulting.com and christi.io. A short script does the scoring. There is no judgment call. The whole run cost $0.37 in API credit. It took about 20 minutes.
This edition has real limits. We state them here. It tested two engines, not five. Each prompt ran once per engine. So one answer can change a small count. We used a US account with no personalization. This edition did not include ChatGPT’s consumer search or Google AI Overviews. It did not include Gemini or Claude. Later editions will add an engine only after a full 28-prompt run on it.
Who do AI engines name for conversational AI consulting?
Perplexity named big firms. Web-grounded GPT named individual people. There were 44 answers to category prompts. The engines named ICX in 5.
Here is the scoreboard for the 22 category prompts. It comes from the published data.
| Engine | ICX named | ICX cited |
|---|---|---|
| Perplexity Sonar | 1 of 22 | 0 of 22 |
| GPT-4o-mini, web search on | 4 of 22 | 3 of 22 |
| Combined | 5 of 44 (11%) | 3 of 44 (7%) |
There were six branded prompts. Both engines named and cited ICX in every answer. That is 12 of 12. An engine that knows the name describes the firm accurately. It links the about, services, and FAQ pages.
Look at the lead prompt. “Who are the best conversational AI consultants for enterprise customer service?” It got two very different answers. Perplexity named eight companies. They were Deloitte, Accenture, Kore.ai, and Cognigy. Then Rasa, Zendesk, Salesforce Agentforce, and Parloa. Web-grounded GPT answered with a list of people. The list started with a conversation designer. There were 22 GPT answers. That designer’s personal site was a source in six. The published JSON holds both answers.
The most cited domains show the same pattern. This is across all 44 category answers. LinkedIn led with 17 source links spread over 8 answers. A freelance marketplace came next. It had 8 links in 2 answers. Then came ICX’s own site with 7. A personal consultant site had 6. A group of chatbot vendors had 4 to 6 each. The counts are source links. One answer can list a domain more than once.
Where did ICX show up? There were two Dallas-Fort Worth prompts. GPT named the firm on both. It cited the site on one. Perplexity named it on one. Perplexity took the name from ICX’s LinkedIn company page. It did not use the ICX site.
On the topical prompts, GPT cited two ICX posts. Then came the best-practices prompt. It cited our post on enterprise chatbot best practices. It listed that post five times in that one answer. It also cited the post on intent architecture once. Perplexity cited no ICX page on category prompts.
Why do individuals beat firms on these prompts?
This result was the surprise in the data. It matters more than ICX’s own score.
A buyer asked “who are the top independent conversational AI consultants.” Both engines answered with people. They did not name companies. Perplexity named three individuals. It said that it kept its list cautious. It gave a reason. The results were clearer on consultants and boutiques. No established global ranking was as clear. GPT listed five people. Three of its five sources were LinkedIn profile pages. The other two were personal sites.
The pattern held across the hiring prompts. Engines that search the live web find LinkedIn profiles. They find personal sites and marketplace listings. Those pages answer the question “who” directly. A firm’s services page answers “what” instead. So engines cite it on branded prompts. They skip it on category prompts.
ICX’s own results confirm the pattern. Some branded prompts asked about the founder. They returned christi.io pages and the ICX about page every time. Category prompts are different. An engine has to find a name, not look it up. Those prompts returned neither page.
What does ICX’s own score say?
Put plainly, ICX is invisible on Perplexity for category prompts. ICX is partly visible on web-grounded GPT. ICX is fully visible once someone already knows the name.
Perplexity named ICX once. The prompt was “Who provides conversational AI consulting in Dallas, Texas?” That answer took the name from ICX’s LinkedIn company page. It described the firm as an IT services company. It cited a competitor’s Dallas landing page. It did not cite the ICX site. GPT cited ICX three times. Two were blog posts. One was the Dallas prompt. Neither engine named ICX on the SaaS, ROI, or services clusters.
That is a 5 of 44 baseline. ICX will measure every later edition against this number.
What should a consultancy do with this?
Four actions follow from the data, in order of weight.
- Lead with the person. On “who” prompts, engines cite LinkedIn and personal sites. Some consultancies hide their people behind a firm name. They give the category prompts away. Whoever appears as a person wins them. The founder’s profile headline must carry the exact words buyers type. So must the about page. The firm’s site must name its author on every post.
- Write definitions and checklists. Definitions and best-practice guides are the pages that engines cite. We had two post citations on category prompts. Both posts define a term or list a checklist. The 80 other posts earned nothing on these prompts. Some pages define a term or give a step-by-step checklist. They are more likely to earn a citation. Short opinion posts are less likely. So we build the next posts in this series that way.
- Fix the entity record. Perplexity described ICX as an IT services company. The LinkedIn company page category said that. An engine that finds a wrong label repeats it. We will correct the company page category and description. We will do it before the next edition.
- Measure, then change one thing. Each edition costs under a dollar and 20 minutes. Run it before and after a change. Then you know if the change worked. A firm that does not run it is guessing. The next edition will test two changes. One is an author entity. One is a Dallas-specific page. Do they raise ICX’s score on local and hiring prompts?
We can run this scoreboard for your brand and prompt set. It is a short project. See ICX services. For the tracking template, see the resources page.
How ICX verified this post
Every number here comes from one run. We collected it on September 12, 2026. We publish the raw answers and the URLs each engine returned. We publish the per-prompt scores. They are available as JSON and as a readable scoreboard. The scoring rule is a fixed text match. It matches the brand names and domains. A script applies the rule. Nobody edits the results by hand.
This post has two limits. Each prompt ran once per engine. So one answer can change a count by one. The engine list has two engines, not five. It did not include ChatGPT’s consumer search. We state both limits. Then readers can compare the next edition honestly.
We draft with AI help. We review every post before it goes live. Read the full policy on the AI disclaimer page. The index runs weekly through October 19, 2026, then monthly. Each edition links back to this one.
Sources
Frequently asked questions
What is the ICX AI Citation Index?
It is a public scoreboard. We update it on a schedule. We run a fixed set of 28 buyer prompts. We use live AI engines. Then we record who gets named or cited. That includes ICX and everyone else. Each edition includes the prompt set and the raw answers. It includes the scoring script. Anyone can check the numbers.
How often will the index be updated?
Weekly through October 19, 2026, then monthly. Each edition uses the same 28 prompts and scoring rule. So readers can compare the trend across editions. We add a new engine only when we can repeat a full run.
Why does ICX publish a score where it mostly loses?
Because the measurement matters most. Most AI visibility advice is opinion. This index is a measurement with a stated method. A buyer or a competitor can run it again. An honest failing score has value. It is worth more than an unverified winning score.
Which AI engines were tested?
Perplexity Sonar and OpenAI's GPT-4o-mini. Sonar searches the web for every answer. GPT-4o-mini had the OpenRouter web search plugin on. We called both engines through the OpenRouter API. Both got the same prompts. Answers had a 700-token cap. This edition did not test ChatGPT's consumer search. It did not test Google AI Overviews.
Why do AI engines name individual consultants more than firms?
Some engines search the live web. They answer a 'who' question with pages that name a person. These pages are LinkedIn profiles, personal sites, and marketplace listings. A firm's services page answers 'what' instead. So engines cite it on branded prompts. They skip it on category prompts. In this run, LinkedIn was the most cited domain. The most named consultant was one person with a personal site.
What does it mean to be cited versus named?
Named means the brand or person is in the answer text. Cited means the answer's sources list a URL from the brand's own site. Cited is the stronger signal. The engine read the brand's page and used it.
Get an email when we publish a new post.