AI Source Index · one AI assistant · 2026-W47
Which websites does Perplexity cite? The most-cited sites, updated weekly
Which websites does Perplexity cite most?
In week 2026-W47, measured 16 November 2026, the websites Perplexity cited most were rtings.com (6 questions, 15%), cnet.com (5 questions, 13%) and forbes.com (4 questions, 10%). Each count is the number of the 40 questions it answered whose answer cited that site at least once.
Week 2026-W47, measured 16 November 2026. One sample per AI assistant per question; a difference of one or two questions is noise, not a trend.
Measured · Maintained by the Promvia team
What stands out
- We ask Perplexity through its Agent API and count the search results its answer cites, not every page it searched.
- This week its 40 answers cited 138 different websites, 4.2 per answer on average (across the 6 AI assistants: 2.1 to 9.0).
- Every one of its 40 answers cited at least one website.
- Among the source types we classify, media sites were its largest group, at 23% of its citations (a domain counts once per answer); 77% matched no type — retailers, vendors’ own sites and the rest.
The 25 most-cited websites
| Domain | Type | Questions | Share (95% interval) | vs week before |
|---|---|---|---|---|
| rtings.com | media | 6 / 40 | 15% (7–29%) | — |
| cnet.com | media | 5 / 40 | 13% (5–26%) | — |
| forbes.com | media | 4 / 40 | 10% (4–23%) | — |
| investopedia.com | other | 3 / 40 | 8% (3–20%) | — |
| pcmag.com | media | 3 / 40 | 8% (3–20%) | — |
| blog.hubspot.com | other | 2 / 40 | 5% (1–17%) | — |
| cnbc.com | media | 2 / 40 | 5% (1–17%) | — |
| europeanpurpose.com | other | 2 / 40 | 5% (1–17%) | — |
| goodhousekeeping.com | other | 2 / 40 | 5% (1–17%) | — |
| lowes.com | other | 2 / 40 | 5% (1–17%) | — |
| mashable.com | media | 2 / 40 | 5% (1–17%) | — |
| nerdwallet.com | media | 2 / 40 | 5% (1–17%) | — |
| nytimes.com | media | 2 / 40 | 5% (1–17%) | — |
| runnersworld.com | other | 2 / 40 | 5% (1–17%) | — |
| shopping.yahoo.com | other | 2 / 40 | 5% (1–17%) | — |
| support.runna.com | other | 2 / 40 | 5% (1–17%) | — |
| wallethub.com | other | 2 / 40 | 5% (1–17%) | — |
| wired.com | media | 2 / 40 | 5% (1–17%) | — |
| wsj.com | media | 2 / 40 | 5% (1–17%) | — |
| accountable.eu | other | 1 / 40 | 3% (0–13%) | — |
| activecampaign.com | other | 1 / 40 | 3% (0–13%) | — |
| activetrail.com | other | 1 / 40 | 3% (0–13%) | — |
| albato.com | other | 1 / 40 | 3% (0–13%) | — |
| androidcentral.com | media | 1 / 40 | 3% (0–13%) | — |
| archestack.com | other | 1 / 40 | 3% (0–13%) | — |
Share = questions whose answer cited the site at least once, out of the 40 this AI assistant answered. The range in brackets is where the true figure very likely falls (a 95% interval); a narrow range means a firm number. “Too few answers” marks a share based on fewer than 32 answers: too few to read as a percentage.
No change column this week: the question list changed, so the two weeks can't be compared.
Method
A fixed, published list of 40 buyer questions in English is put once to each of 6 AI assistants every Monday. Every URL an answer cites is stored, and a domain counts once per question per assistant. One sample per assistant per question.
How we measure Every observation behind 2026-W47, as JSON (this AI assistant’s rows carry "engine": "perplexity")
For the AI Source Index we publish every observation behind the weekly tables, row by row, so anyone can recount them.