The websites behind repeated AI recommendations

Data rechecked 20 October 2026. We removed one group's data and recounted every figure.

In our app-development research, 3 Sided Cube appeared as a supplier candidate in 52% of Claude’s answers. In the subset of answers that did not cite its website, it appeared in 15%. The same five companies still led the sample, but their order changed.

That difference matters to a marketing team reporting how often its company makes an AI shortlist. Repeated appearances can come with repeated sources. A recommendation count alone does not show how much of that visibility sits alongside the same website.

Across our technology, cyber security and app research, those recurring websites included suppliers’ own sites, competitors’ sites, a sector directory and comparison publishers. The pattern varied by market and engine.

What we studied

We reviewed 311 saved answers to 104 research-written questions; 305 had saved sources. The records came from existing company and prospect-market research, collected between 30 July and 21 August 2026. They are not transcripts of real buyer conversations or a benchmark of today’s engines.

The three reporting groups are Technology, Cyber security and App development. Technology covers IT support.

We counted a recommendation when an answer affirmatively offered a named supplier or solution for the requested purchase, including conditional alternatives. This is shortlist inclusion, not necessarily an explicit first choice. We then measured which websites appeared alongside those recommendations.

App development: the same leading companies, in a different order

Claude’s app answers covered charity, public-sector and social-impact buying situations. In the complete sample, 3 Sided Cube was the most frequent candidate at 52%, followed by Reason Digital at 42%.

We then set aside every answer citing 3sidedcube.com. That left 56% of the original sample. Reason Digital became the most frequent candidate at 26%; 3 Sided Cube moved to fifth at 15%.

Reason Digital’s share also fell. It became the relative leader because the other frequencies changed, not because its visibility increased. Appoly, Red C and Torchbox tied for second place in the remaining answers. Rank here means frequency across answers, not the order of names within an answer.

A recurring website was usually accompanied by other sources

Every ChatGPT / Codex app answer recommending 3 Sided Cube also cited 3sidedcube.com. None used that website as its only saved source. Across those recommendation answers, the saved source lists contained 51 distinct website addresses.

Claude showed the same association less consistently: 84% of its 3 Sided Cube recommendation answers cited the company’s website; 4% cited it alone. Gemini’s sources were more dispersed, with four websites tied at 38%. None of those tied leaders was 3 Sided Cube’s own website.

A competitor’s website also appeared repeatedly. Every Gemini answer recommending ANTLR Interactive or Goji Labs included simpalm.com in its sources. Simpalm is another app supplier. These records retain website names rather than exact pages, so we cannot establish which page was used or whether it supported a particular recommendation.

Technology: supplier websites recurred alongside other sources

In the IT-services questions, Grapevine appeared in 47% of Claude’s answers. Its website accompanied 81% of those recommendations, always alongside other sources. In the subset that did not cite grapevine.uk.com, Grapevine appeared in 14% of answers. SES Computers and Rejuvenate IT became joint leaders, each at 36%.

The same exclusion affected other suppliers too. HGC IT moved from 27% to 11%, although the excluded website belonged to Grapevine. Removing a whole answer can remove several recommendations together. This is why the result cannot be read as the isolated effect of one company’s website.

Cyber security: a shared directory accompanied many suppliers

CREST Marketplace appeared in 91% of Claude’s cyber-security answers. Every Claude answer recommending NCC Group, LRQA, Pen Test Partners or Prism Infosec included that directory. Each also included other sources.

Those websites serve different purposes. CREST Marketplace is an accredited-provider directory. NCSC is the National Cyber Security Centre. DeepStrike is a security supplier whose website publishes provider comparisons. Counting their citations does not make them equivalent forms of endorsement or assurance.

Excluding Claude’s CREST-citing answers left just one answer. A new ranking would mostly describe that one question. The useful result is how widely the directory appeared across the saved answers; there is too little remaining evidence to say which suppliers would lead without it.

What this adds to an AI visibility report

For tech, cyber security and app companies, the research adds context to the recommendation count. A company can appear repeatedly with its own website, a directory shared with rivals, or a publisher that compares the market. Those are different source patterns, even when the headline visibility percentage looks similar.

Our results support reporting the recurring websites alongside the questions and recommendations. They also give marketing teams a reason to investigate relevant sources beyond their own website. They do not establish that getting a directory listing or publishing a comparison will increase recommendations. That requires a separate test.

Avouch can use this research approach to review where your company and competitors appear, then agree the relevant work: website and content improvements, industry profiles, publication outreach or community activity. Delivery packages have an agreed scope and outputs; the source pattern helps inform the work, without promising inclusion in AI answers.