Proxy sellers wrote 59% of what Google's AI cites about proxies

On 8 August 2026 we captured every source Google's AI Overview cited across 15 proxy queries, in a real browser, expanding each overview through Show more and Show all sources. That produced 162 citations from 67 distinct hosts. Proxy sellers wrote 59.3% of them.

Who Google is quoting

The first citation slot is the one a summary leans on hardest, and a company that sells proxies holds it on thirteen of fifteen queries.

The YouTube row is not a separate category, and reading it as one understates the finding. We went through the video descriptions on the wider 26-query capture: of 45 YouTube citations, 25 sit on channels owned by proxy vendors, and 17 of the remaining 20 carry affiliate rails to proxy vendors. That is 42 of 45, 93%, vendor-owned or vendor-funded. Three slots are neither: one channel with no rails visible, one belonging to the author of SeleniumBase, and one off-topic misfire.

So the 59.3% in the table is the floor, not the ceiling. Add the video slice and the share of citations written or paid for by somebody selling proxies is higher than any single row shows.

These are the fifteen, so you can run them yourself and see whether the mix has moved. Every one returned an overview, and the mean was 10.8 citations:

best residential proxies            bright data alternatives
no kyc proxies                      oxylabs alternatives
cheap residential proxies           isp proxies vs residential proxies
bright data vs oxylabs              best proxies for web scraping
decodo vs oxylabs                   best proxies for cloudflare
iproyal vs webshare                 are residential proxies legal
residential proxy pricing           how do residential proxies work
proxy free trial

A study that will not name its queries cannot be checked, and this one is asking you to check everybody else.

The clearest single example is the query bright data vs oxylabs. Its overview is sourced first to a page on Oxylabs' own site about why you might leave Bright Data, and second to Bright Data's mirror image of the same argument. Two competitors grading each other, summarised as a neutral answer. Both pages still say so in their own meta descriptions, quoted in the sources below.

None of that is against the rules. A vendor publishing a comparison is doing marketing, which is allowed and expected. The problem is what happens when a summary layer strips the byline off and presents it as consensus.

Not one of them carried a date

This is the part that should change how you read any proxy price you did not fetch yourself.

Zero of the 15 overviews carried an as-of date, a verified timestamp, or a prices-vary hedge, while asserting hard figures. Sampled from the body text:

residential proxy pricing gave a range of $1.00 to $15.00 per GB, with DataImpulse at $1.00/GB, Webshare at $1.40 to $3.50/GB, IPRoyal at $7.00/GB and Oxylabs moving from $4.00 to $2.50/GB. No date on any of them.

The only caveat anywhere was Google's own boilerplate that AI can make mistakes.

Proxy pricing moves weekly. A provider's whole residential line can change shape inside a fortnight, and one of the providers here replaced its pricing model outright in the weeks before we ran this. A confident, undated price in a market like this is worth very little, and the summary layer does not carry the timestamp even when the page it read had one. The captcha solving market has the same shape and the same gap: the one review site covering it properly prints no read date at all, which is half of what a captcha solve actually costs.

Reddit is not the escape hatch

The usual advice is to ignore the review sites and ask a forum. Reddit does show up: it is in the top five results on 20 of the 26 money queries we measured.

The catch is age. The median Reddit thread ranking on those queries is 511 days old and carries 26 points. In a market that reprices weekly, that is a consensus formed a year and a half ago, still being served as current. Age also detaches a thread from the company it describes: the provider now called Decodo spent those years trading as Smartproxy, so the highest-ranking discussion of it is filed under a name it no longer uses.

And it is being worked. In a separate corpus we tuned to find outcome reports, 253 of 995 threads were affiliate or vendor-run, which is 25.4%. One account supplied 84 of them.

Disclosure

We are not outside the incentive structure. This site will earn affiliate commission once those links are switched on, the rates differ between providers by as much as five times, and that is a permanent conflict rather than one we have solved.

What we do about it is narrow and checkable:

• Rankings use data only. Commission is not an input to any sort or filter on this site, and a build check enforces that rather than a promise.
• Every figure carries the date we read it, and a link to the page we read it on, so you can check the number rather than trust us.
• Where a figure cannot be produced honestly, we publish a dash and the reason instead of a plausible number.
• We do not declare winners. We publish no benchmark data, so we do not rank providers on quality, and we say so rather than implying one.

That is a weaker claim than the one most comparison sites make, and it is the one we can actually support. The test to apply to us is the same one this post applies to everyone else: find the number, click the source, check the date. Our method and our commercial arrangement are both written down, and the table shows its working.