Who Do AI Engines Actually Cite? What 74 Tracked Buyer Queries Show (July 2026)

TL;DR: Three Findings From 74 Tracked Queries

I track 74 commercial buyer queries against AI engines every week, because my business depends on whether those engines mention me. Three findings, as of July 31, 2026:

1. The engines disagree about who to trust. My site is cited on 14 Perplexity queries and has zero mentions in ChatGPT's citation dataset. Same content, same week, opposite outcomes.
2. Reddit is top-three on both engines. 110 mentions in ChatGPT's "ai consultant" space, 5,963 in bookkeeping, and 5 appearances across my Perplexity set. It is the only source both engines lean on.
3. Unknown sites regularly beat big brands. Sites you have never heard of tie or outrank Forbes on both engines. Citation is not a popularity contest, which is the most useful thing in this entire report.

Everything below is my own tracking data with the methodology and the sample sizes attached. Nothing here is a vendor study.

What I Track, and Why I Bother

In mid-July 2026 I started logging which sources AI engines cite when someone asks a buying question in my market. It started at 24 queries. It is 74 now.

The queries are all things a real buyer types before spending money — questions to ask before hiring an AI consultant, how much does AI consulting cost, best AI visibility tools, plus industry-specific versions for accounting, insurance, real estate, dental, and property management. Not vanity keywords. Purchase-intent questions.

Two instruments, because one is not enough:

• Perplexity — checked weekly through its API against my exact query list, about a penny per query. This tells me whether my specific pages get cited. (I wrote up how I built that checker, and there are paid tools that do it with dashboards if you'd rather not.)
• ChatGPT — measured through DataForSEO's LLM-mentions dataset, which reports which domains get cited across an entire prompt space. Different shape of data: it tells me who is winning rather than whether I am.

Finding 1: What Perplexity Cites

At the July 27, 2026 checkpoint, 14 of 62 tracked queries returned a citation to my site — 23%. Here is who was taking the rest, counted by how many of those 62 queries each domain appeared in:

Read the bottom half of that table twice. Four sites you have never heard of are tied with each other and within one appearance of Forbes. Perplexity is not ranking by brand recognition — it is ranking by whether a page answers the exact question asked, in a format it can lift.

And the number one source is a video platform. YouTube went from 8 appearances across 24 queries on July 15 to 16 across 62 on July 27 — it doubled its absolute lead in twelve days. If you have been treating video as a separate marketing channel from search, that gap is the reason to stop.

Finding 2: What ChatGPT Cites

Different instrument, different picture entirely. Here is the "ai consultant" prompt space — 405 tracked mentions against 14,681 monthly AI search volume, US English, pulled July 31, 2026:

Zero expert blogs. Zero consultancies. Zero personal sites. Press, a forum, government statistics, research papers, and job boards. That is a completely different trust model from Perplexity's.

The bookkeeping space is bigger and tells the same story louder — 24,130 mentions against 1,148,376 monthly AI search volume:

Reddit holds a quarter of the entire cited space by itself. 5,963 of 24,130 mentions — three times Indeed, nearly four times Forbes. And legalclarity.org, a site with no brand, no press coverage, and no reputation you could name, holds 550 mentions inside a space that Forbes holds 1,585 in. It is playing in the same league as Forbes on a fraction of the authority.

Finding 3: The Engines Disagree, and It Costs You

Here is the finding I did not expect and the one that changed how I work:

I have been publishing answer-shaped expert content for months. On Perplexity it works — 14 citations. In ChatGPT's citation dataset, that same body of work does not exist. A win on one engine does not transfer to the other.

The last row of that table is the exception worth knowing, and I have direct evidence for it. My other business, an agency, gets cited by ChatGPT on exactly ten prompts — and all ten are commercial cost questions: "best AI phone answering service" (26 monthly AI searches), "how much should social media marketing cost" (25), "how much does a virtual agent cost" (21), "AI call center agent cost" (14), "content agency cost" (14). No thought pieces. No frameworks. Pages with prices on them.

So ChatGPT is not uniformly closed to businesses. It is closed to businesses explaining things, and open to businesses publishing what they charge.

What This Means If You Want AI Engines to Cite You

Four things, in the order I would do them:

1. Publish answer-shaped expert pages — this wins Perplexity. The question as the heading, the answer in the first paragraph, specific numbers, a named author with real operating history. This is the format that lets dancumberlandlabs.com tie Forbes. You do not need authority; you need to be the most direct answer to a specific question. The full mechanics of answer engine optimization are their own topic.
2. Publish your prices — this is the one ChatGPT door that opens for a business. Every page of mine ChatGPT cites is a "how much does X cost" page with real numbers. Most companies won't do this, which is exactly why it works.
3. Show up on Reddit like a person. It is the only source both engines rank in their top three, and it holds 5,963 of 24,130 mentions in one space I measured. Answer questions where your buyers already are, honestly, without the pitch. Being useful is the only version that survives moderation, which is the reason the channel still has value.
4. Publish original data. This post is the play. Numbers nobody else has are the one thing press and language models both have a reason to cite, and they compound — a study gets cited for years while an opinion piece gets cited for a week. If you track anything about your own business, you already have the raw material.

Do this today: pick the five questions your buyers actually ask before hiring you. Paste each one into Perplexity and ChatGPT. Write down who gets cited. That is your competitive set in AI search, and I promise it is not the list you expected.

Methodology (Read the Caveats — They're the Point)

Original data is only worth citing if you can see how it was made. So:

• Sample size. 74 tracked buyer queries as of July 31, 2026, grown from 24 in mid-July. The citation-rate figure (14 cited, 23%) is from the July 27 checkpoint, when the set was 62 queries. It is not yet recalculated against 74.
• Denominator warning. The rate went 25% (14/57) on July 22 to 23% (14/62) on July 27 — because I added queries, not because I lost citations. The absolute count was flat at 14. This is why I publish the raw count next to the percentage.
• Engines covered. Perplexity and ChatGPT only. No Google AI Overviews, no Claude, no Gemini, no Copilot. Do not extrapolate this to engines I did not measure.
• Two different instruments. The Perplexity data is my own query list checked directly through the API — it measures my pages. The ChatGPT data is DataForSEO's LLM-mentions dataset, which measures a whole prompt space, not my query list. The two columns are not strictly comparable, and I am not going to pretend otherwise.
• Locale. United States, English. ChatGPT figures pulled July 31, 2026; the dataset moves week to week (my own pull four days earlier gave Business Insider 111 rather than 131).
• Cadence. Perplexity weekly, ChatGPT at least monthly.
• Bias disclosure. I sell AI consulting. These queries are my market, which means my query set is not a neutral sample of the internet — it is a neutral sample of my buyers' questions. Take the patterns, not the specific domains.

This Report Updates Monthly

Next update: late August 2026, with the full 74-query set recalculated and a longer trend line on whether YouTube keeps climbing. If a finding here reverses, I will say so on this page rather than quietly editing the number.

If you want the update when it lands — plus the weekly version of what I'm testing, what it cost, and what broke — that's my newsletter. Join The AI Leader's Weekly Playbook: one email a week documenting how a one-person operation actually runs on AI, numbers included. No course funnel at the end of it.

If you'd rather skip ahead to your own numbers, the $2,500 AI Assessment includes a visibility baseline across the engines for your business specifically — which prompts you appear on, who holds the ones you don't, and what it would take to take them.

Related reading: answer engine optimization, the best AI visibility tools, how to track brand mentions in AI, what an AI audit actually covers.