What AI Engines Actually Cite When Buyers Ask Who To Hire

A no-answer is an engine returning no answer text for a question. We exclude those before counting, because an engine that did not answer cannot have cited anything. In this run there were none, so the exclusion changed nothing. Answers that returned text but no source list are kept and counted, and section 06 reports them.

The question set is our own: 103 questions a buyer would actually type when looking for an AEO, GEO or AI visibility provider, spread across nine segments. Two shares get reported for most figures. Share of answers uses a denominator of 514, one question on one engine. Share of cited links uses a denominator of 2,955, every individual source link the engines returned.

Not one review directory appears in the top 25. Our own domains, sourcerankai.com and danishleadco.io, are left out of this ranking. Their links are counted in every total on this page, but we do not publish our own citation figures on our own question set.

Community is Reddit and Quora. Social and video is LinkedIn, YouTube, Facebook, Medium and Instagram. Review directory is the nine sites listed in section 03, four of which appeared. News is Yahoo, Search Engine Land, Forbes, Business Wire, Search Engine Journal, PR Newswire, Adweek, Entrepreneur and Fast Company. Docs and standards is Google developer and support documentation, OpenAI documentation, schema.org, arXiv and government domains such as nih.gov, fda.gov and sec.gov. Vendor or agency site is the residual: 1,329 commercial sites belonging to agencies, tools and software vendors.

Counted at the registrable domain, so content subdomains roll into their parent. One further directory, getapp.com, is in the rule set and was not cited in this run, so the wider nine-site definition gives the same 12 links and 11 answers.

Against a denominator of 514 answers, the eight named directories are 2.1%. Against a denominator of 2,955 cited links, they are 0.4%. GoodFirms, Sortlist, Trustpilot and TrustRadius were not cited a single time in any of the 514 answers.

This does not mean directory profiles are worthless. They still get human traffic, and a profile can feed the reviews and third party mentions an engine reads elsewhere. It does mean that on this question set, on this date, an AI answer engine deciding who to recommend almost never reached for one. Our own read of what actually earns the citation is on the AEO agency comparison.

The combined row counts each answer once even when it cited two of the three, which is why 126 is less than 59 plus 49 plus 48.

Three consumer platforms were cited in 126 of 514 answers and touched 76 of the 103 questions. LinkedIn was the single most cited domain in the entire run, and Reddit the second. An engine synthesising a hiring recommendation is reading what people said in public, on a forum, on a profile, on video, far more than it is reading a structured vendor listing.

Half the citations come from 190 different places. Two thirds of the domains that got cited got cited once and never again. That shape argues against any strategy built on placing yourself on a handful of high authority pages, and in favour of being consistently citable across a wide surface, because there is no small set of sources to buy your way onto.

30 of 514 answers, 5.8%, carried no source list: 17 from ChatGPT, 11 from Gemini and 2 from Google AI Overviews. In the 17 September run it was 120 of 500, because Gemini attached a source list to none of its answers that day. The count moves with what each engine exposes and what our collector could capture on the day, which is not evidence that any engine invented its answers.

This matters for anyone quoting a single run, including us. Every domain share on this page is computed over the 484 answers that did carry sources, then expressed against the full 514 so the reader can see the real denominator. A vendor who quietly drops the sourceless answers from the denominator will report the same run with higher shares. If an AI visibility statistic does not tell you how many of its answers had no sources, you cannot tell which denominator it used, and the number is not checkable.

In the logistics segment, docs and standards and social and video tied on 10 links each; docs and standards is shown as runner up. Segment link counts sum to 2,955 and segment question counts sum to 103.

Vendor and agency sites are the most cited kind in all nine segments, never below 80% of that segment's links: from 81.7% in healthtech to 96.4% in agencies. What moves is second place. In five of the nine segments, general AEO buying, B2B SaaS, professional services, financial services and home and commercial services, it is social and video, which in practice means LinkedIn, YouTube and Facebook. In healthtech and manufacturing it is documentation and standards: Google's own developer docs, schema.org, government sources. In logistics the two tie, and in the agencies segment second place is community, meaning Reddit. In a technical or health vertical the engine reaches for a specification more often than a post. We run the same cut across eleven companies in eleven different industries, where the swing is far larger than it is between segments of one question set: the source mix by industry.

• +Review directories were cited in 11 of 514 answers, 2.1%, and in 12 of 2,955 links, 0.4%. GoodFirms, Sortlist, Trustpilot and TrustRadius were never cited.
• +Reddit, LinkedIn and YouTube were cited in 126 of 514 answers, 24.5%, touching 76 of the 103 questions. LinkedIn was the single most cited domain.
• +Vendor and agency websites supplied 2,522 of 2,955 links, 85.3%, and were the most cited kind in all nine segments.
• +Citations are long tailed: 1,364 domains, 190 needed to reach the first half of all links, 910 cited exactly once.
• +30 of 514 answers carried no source list, 17 of them from ChatGPT, which is why the denominator has to be published alongside any AI citation statistic.