Do AI Engines Recommend the Agencies That Rank on Google? 2,970 AI Answers (2026 Study)

So the AI shortlist and the Google results page for the same agency search are, for the most part, two different lists. The three engines agree on that to within a point. Widen the bar to Google's top 20 and the share still stays under 15% on every engine.

Agencies get named about 3.4 to 3.7 times as often on the questions where they rank as tracked agencies do on the questions where they do not. Inside a single agency, that gap shrinks to roughly half, and a few agencies carry most of it. Our read is that brand size fits the half between agencies, and that the half inside one agency cannot be pinned on ranking.

If your question is the narrower one, whether ChatGPT pulls its answers from Google at all, we answered that separately in where ChatGPT gets its information.

How we ran the study

We track buyer questions for LoudFace in Peec AI, which runs each question on ChatGPT, Perplexity and Google AI Overviews every day and records the answer, the sources it cites and the brands it names. We wrote the method down before we read a single answer or results page, then amended it three times: to read the real site behind masked Google links, to leave out one generic brand name and add a check of whether the gap holds inside one agency, and, after review, to measure the size of that check.

The questions. We picked them by a fixed rule: every agency question we actively track in Peec that is linked to at least one Google keyword containing "agency", "agencies", "consultant", "consultants" or "services". That gave 33 questions, linked to 39 Google keywords (1 to 3 per question). A few examples: "Best AEO agencies 2026" with the keyword "aeo agency", "Best SEO agency for B2B SaaS companies" with "b2b saas seo agency", and "SEO agency for fintech companies" with "fintech seo agency". The pairing of question to keyword already existed in our records. We did not choose it for this study.

The answers. Every answer Peec recorded for those 33 questions from 6 October to 4 November 2026, 30 days, one per question, engine and day. That is 2,970 answers: 990 per engine and 90 per question. Google AI Overviews returned no answer text on 22 of its 990 records. They count in every per-answer rate, which makes those Google AI Overviews rates slightly low. They add no namings and no cited sources, so the shares of namings and of cited sources are unaffected.

The Google results. We pulled Google's organic results for each of the 39 keywords on 5 November 2026: United States, English, desktop, top 20, through DataForSEO. In the first pull Google returned 557 of 736 results as masked redirect links, so we pulled the pages again the same day and read each result's real site from its breadcrumb (the second pull returned 737 results, and every one resolved).

What counts as named. An agency is named when Peec's brand detection lists it in the answer, the same unit behind Peec's visibility metric, which "measures the percentage of AI responses that mention your brand." Peec tracks 53 brands in our project, each by a tracked name and its own domains. We left out two: LoudFace, which publishes this study, and GEO Agency, whose name is a generic phrase the detector cannot tell apart from ordinary words. That leaves 51 tracked agencies.

What counts as ranking. An agency ranks for a question when one of its own domains holds a top-10 (or top-20) organic result for any keyword linked to that question.

Limits, up front. Agencies Peec does not track are invisible to the counting rule, so a firm that gets named a lot but sits outside our tracked set never shows up. A results page is one day's snapshot and can move the next. The Google side is US desktop only. Peec runs the question text, while Google ranks the keyword, so the two sides are matched through our own question-to-keyword links. We kept the raw answers and results pages, along with the scripts behind every number here.

About 1 in 10 times an engine named an agency, that agency ranked in Google's top 10 for the same search

The headline holds across all three engines and under every check we ran.

The intervals in the first table come from resampling whole questions 2,000 times. They are wide, because 33 questions is a modest set and a few questions carry a lot of the ranking agencies. The upper ends still sit around 16% on every engine.

We tried three other cuts:

• One keyword per question. Using only each question's first linked keyword: ChatGPT 9.0%, Perplexity 9.2%, Google AI Overviews 9.4%.
• Only questions where someone ranks. On 27 of the 33 questions at least one tracked agency had a top-10 result. Restricted to those 27: ChatGPT 11.5% (5.8% to 19.4%), Perplexity 11.2% (6.2% to 19.3%), Google AI Overviews 12.5% (7.4% to 19.4%).
• Question by question. How many questions had more than half of their agency namings go to an agency ranking top 10? ChatGPT 2 of 33, Perplexity 3, Google AI Overviews 1.

On almost every agency question we track, the engines mostly name firms whose sites you would not find on page one of Google for that search.

Google AI Overviews sits on the same page as the organic results, and it still does not lean much harder on the agencies ranked below it. Its 10.3% is within a point of ChatGPT and Perplexity.

Agencies are named 3.4 to 3.7 times as often on the questions where they rank. Is ranking the reason?

Our read: about half of the gap is between agencies, which fits brand size; the half inside one agency, this data cannot pin on ranking.

Here is the comparison that looks like a yes. We paired each of the 51 tracked agencies with each of the 33 questions: 1,683 agency-question pairs. In 51 of those pairs the agency ranks top 10 for that question's search. Then we compared how often the agency was named in the two groups.

That table compares different agencies, though. An agency that ranks on some questions may also be a bigger, better-known firm, so the next check holds the agency fixed.

Inside one agency, the gap shrinks to roughly half. Take the 22 tracked agencies that rank top 10 for at least one question. For each, compare how often it is named on the questions where it ranks against the questions where it does not. By direction, they split close to even:

• ChatGPT: 11 named more often where they rank, 10 less often, 1 tied.
• Perplexity: 9 more often, 10 less often, 3 tied.
• Google AI Overviews: 12 more often, 10 less often.

By size, the gap is still there, but smaller. Averaged over the 22 agencies:

Inside one agency the difference is roughly half the gap between agencies. Perplexity's interval crosses zero, and the medians are small. On ChatGPT 12 of the 22 agencies sit within 5 points either way, and on Perplexity 14 of 22. Google AI Overviews spreads wider: 8 of 22 within 5 points, 10 more than 5 points above and 4 more than 5 points below. A few large differences carry the average. The three largest, up to +59.4 points, carry 78% of the summed differences on ChatGPT and 61% on Google AI Overviews. On Perplexity they carry more than the whole net difference, because the other agencies net out below zero.

Our read, which we did not measure directly: bigger, better-known agencies may rank for more of these searches and get named more often everywhere, which fits the half of the gap in the first table that sits between agencies. Holding the agency fixed holds its brand fixed, so the part that remains inside one agency is not brand size. It could be ranking, or the agency's fit to the question (an agency that ranks for a vertical search may specialise in that vertical), and this data cannot tell those apart. So the study makes no causal claim either way: it does not show that ranking causes naming, and it does not rule out a modest effect.

Which agencies do AI engines name most, and where do their sites rank?

Share of the 990 answers per engine naming each agency, with the number of the 33 questions where its own site ranks top 10. The table is ordered by total namings across the three engines combined, so the order on any single engine differs:

Four of these eight agencies have no top-10 result for any of the 33 searches. iPullRank and Skale are not even in the top 20 for any of them, and Google AI Overviews names Skale in 24.9% of its answers anyway.

Question by question it looks the same. On 25 of the 33 questions, the agency named most across the three engines combined had no top-10 result for that question's search. The 8 where it did:

• "Best AEO agencies 2026" (First Page Sage)
• "Best AEO agency for B2B SaaS 2026" (Omniscient Digital)
• "Best SEO agency for B2B SaaS companies" (Omniscient Digital)
• "Top AEO agency for fintech 2026" (First Page Sage)
• "SEO agency for cybersecurity SaaS companies" (First Page Sage)
• "SEO agency for fintech companies" (Siege Media)
• "Which agencies do SEO and AI search for security and compliance software?" (First Page Sage)
• "CRO agency that understands AI search traffic for B2B SaaS" (Omniscient Digital)

First Page Sage is the case that looks like ranking works

First Page Sage ranks top 10 on 7 questions, the most of all 51 tracked agencies (next are Discovered Labs and Revv Growth on 5), and its own list of AEO agencies sits at position 1 on Google for "aeo agency" (5 November 2026). On those 7 questions it is named far more often than elsewhere: ChatGPT 62.4% against 25.4%, Google AI Overviews 57.6% against 21.3%, Perplexity 36.7% against 16.0%. Omniscient Digital shows the same shape on its 4 questions (ChatGPT 92.5% against 55.3%).

Note what is ranking there. For "aeo agency" it is a list page, the kind of page engines read and quote.

Then the mixed cases, where the picture changes by engine. Directive Consulting ranks top 10 on 1 question. There ChatGPT names it in 33.3% of answers against 25.2% elsewhere, while Perplexity names it in 0% against 14.7% and Google AI Overviews in 6.7% against 22.3%. TripleDart ranks on 3 questions: ChatGPT names it in 1.1% of answers on those against 10.9% elsewhere, Perplexity 0% against 3.9%, and Google AI Overviews 8.9% against 6.2%. Cases like these sit on both sides of the direction count. The positive average comes from a few large differences.

What holds Google's top 10 for agency searches?

Mostly not the agencies the engines name. We typed the 330 top-10 results for each question's first keyword:

The most frequent hosts in that last group were krowdbase.com (9 results) and revenuebase.ai (8). So for these 33 searches (one day, US desktop), Google's first page is mostly other sites, from agencies we do not track to publishers and lists, with social threads and directories mixed in. The 51 agencies we track hold 52 of the 330 slots; agencies we do not track are invisible to this count.

How often do the engines cite the sites Google ranks?

Rarely, and ChatGPT least. Share of each engine's cited sources whose domain holds a top-10 result for the question's search:

We measured this by domain because Google masked most result links in the raw data and a masked link does not tell you the exact page. A domain match is the looser test, so if anything these shares run high.

The order matches what we have seen elsewhere: each engine draws on a different pool of pages. In earlier work we found Google AI Overviews leans on YouTube, LinkedIn and Reddit, and ChatGPT on lists, arXiv papers and Reddit. ChatGPT's 4.1% fits that. Seer Interactive's study is closer context: across about 100 questions, more than 87% of ChatGPT search citations matched Bing's top 20 organic results, against a 56% match for Google. OpenAI says ChatGPT search "sometimes partners with other search providers" and "typically rewrites your query into one or more targeted queries that it sends those providers" (OpenAI Help Center). The rewritten searches are not the keyword you typed into Google, so there is little reason to expect the same top 10.

How this compares with published studies

Several studies have measured how much AI answers overlap with Google's results. Every one we found measures cited pages or domains. None measures which provider an answer recommends, which is the unit here. Keep the unit column in view before comparing numbers.

The comparison worth making is with Ahrefs. Its March 2026 update found 37.9% of the URLs cited in AI Overviews also appeared in the first 10 result blocks, SERP features included (37.1% counting blue links only), down from about 76% in July 2025, which Ahrefs reads as "Google is selecting far fewer pages straight from the original SERP". Our domain-level figure for Google AI Overviews is 12.0% of cited sources. The two are not the same measure (a URL per keyword against a domain per linked keyword, and Peec runs the question text rather than the keyword), so we report both and do not try to reconcile them.

Why would an engine name an agency that does not rank?

Our read: because it reads the agency's name on a page it retrieved, and for agency questions that page is often a list. What follows draws on this study and on earlier work.

Google's own guidance says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It also says both features may use "query fan-out", "issuing multiple related searches across subtopics and data sources" (Google Search Central). An answer to "best SEO agency for fintech" is built from several searches. For these agency keywords, Google's own top 10 mixes list pages, social threads and directories with agencies outside our tracked set; tracked agencies hold 52 of the 330 slots.

Our earlier studies point the same way:

• In an earlier, smaller read of 120 answers, the engines named LoudFace only when they had read a loudface.co page. The fuller study, AI engines name the brands they read, found the same with few exceptions: LoudFace was named 631 times, and 624 of those answers (98.9%) had read a loudface.co page.
• The engines cite agency list pages, ours included, and name the agencies inside them.
• In earlier work, Domain Rating did not decide who gets cited for agency-list questions.

Our read from those earlier studies, not something this study measured: the engines name the agencies that appear on the pages they read. Sometimes that page is the agency's own and it also ranks. The First Page Sage case suggests it is often a list page; this study did not measure how often. In this study, iPullRank and Skale, with no top-20 result for any of the 33 searches, were named in 12.3% to 25.5% of answers on each engine.

What it means if you are buying or selling agency services

None of what follows was tested by the study. It is our read of the data.

If you are choosing an agency

1. Treat the AI shortlist and the Google results page as two different lists. On 25 of 33 questions the agency named most across the three engines combined had no top-10 result. Check both before you decide who to call.
2. Ask more than one engine. Skale was named in 24.9% of Google AI Overviews answers and 12.3% of Perplexity answers. One engine's shortlist is one sample.
3. Open the page behind the name. If an engine names an agency because a list named it, the list's criteria are what you are actually relying on. Read them.

We have run other cuts of the same kind of data: SEO versus AI search agencies, how the question is phrased and whether your industry changes the answer.

If you want AI engines to name your agency

Our inference: ranking your own site for "your-niche agency" goes with a lift inside one agency of about 5 to 8 points on average, uneven across agencies and carried mostly by a few. The gap between agencies is much bigger.

Our earlier studies point to another route: being on the pages engines read. First Page Sage's best result was its own list at position 1. Agencies with no top-20 result at all were named in about a quarter of one engine's answers. If you want to be in the answer, work on the lists and third-party pages that already get cited for your questions. We covered how in getting into the best-of lists AI answers pull from, and our GEO agency page covers the service side.

Do not drop Google to do it. This study measured AI answers, not the buyers Google sends you. Our point is narrower: a ranking on your own site and a place in the AI shortlist are separate goals. In this data the first went with a lift of about 5 to 8 points in the second inside one agency, uneven across agencies and much smaller than the gap between agencies, so work on both.

Limitations

• The tracked set. The counting rule sees 51 agencies. A firm Peec does not track for us is invisible, whether it ranks or not.
• 33 questions. The intervals are wide, and a handful of questions carry many of the ranking agencies.
• One day of Google. The results pages are from 5 November 2026. A results page can move the next day. We do not claim a stable rank.
• US desktop only. Other countries, mobile results and personalised results may differ.
• Matching. Peec runs the question text and Google ranks the keyword. We matched them through our own question-to-keyword links, 1 to 3 keywords per question.
• Domain level. The citation overlap is measured by domain, because Google masked most result links.
• No cause either way. The study does not show that ranking causes naming, or that it does not.