Empire325 Research · July 2026
The State of AI Agency Recommendations 2026
We asked three AI answer engines — Claude, ChatGPT, and Perplexity — to name the best marketing agency across 16 buyer verticals, then recorded which sources each one cited, and corroborated it with a live search snapshot. The headline: the recommendation set is concentrated but engine-specific — no single source wins across all three. Every figure below traces to a dated raw-data file. Free, citable, CC BY 4.0.
By Milton James Acosta III, Founder & CEO · Published 2026-07-22 · Last reviewed 2026-07
TL;DR: When AI recommends a marketing agency, the shortlist is concentrated but each engine has a different favorite. In Claude (June 12), firstpagesage.com anchors 5 of 16 verticals; in Perplexity (June 13) the most-cited source is reddit.com at 11 of 16; in ChatGPT (June 13) dataally.ai leads at 5 of 16. No single source wins across all three. And 74.7% of Claude's supporting citations were listicles, not agency homepages — the recommendation layer is won off-site, in the roundups and community threads AI actually reads.
Key findings · cite this
The headline numbers, ready to quote
Free to cite under CC BY 4.0 with attribution to Empire325 Marketing and a link to this page. Copy any stat below, or download the full aggregated dataset as CSV or JSON.
- ▸
74.7% of the pages Claude cited to justify a “best agency” recommendation were listicle/comparison articles, not agency homepages (115 of 154 classified cited pages, 2026-06-12 baseline). (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
Perplexity cited reddit.com in 11 of 16 “best [vertical] marketing agency” verticals on 2026-06-13 — its single most-cited source. (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
In Claude, just two domains — firstpagesage.com (5 of 16 verticals) and thriveagency.com (4 of 16) — anchored the largest share of agency recommendations (2026-06-12). (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
In ChatGPT, dataally.ai led agency recommendations across 5 of 16 verticals on 2026-06-13 — a different anchor domain than Claude’s or Perplexity’s. (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
ChatGPT cited empire325marketing.com/industries/asset-management in-model on all 3 tracked dates while the same page ranked in 0 live organic SERP results — AI citation and Google ranking are distinct signals. (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
124 unique domains supplied Claude’s answers across 16 buyer queries, yet Empire325 appeared in 0 of them (2026-06-12 baseline). (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
Roughly 11 domains recur in both the June in-model AI tracker and the July 22 live organic SERP, confirming a shared listicle-and-directory citation layer across methods. (Empire325 Marketing, The State of AI Agency Recommendations 2026) - ▸
NoGood (nogood.io) surfaced in 4 of 9 live buyer-query SERPs on 2026-07-22 — the most pervasive single agency in that snapshot. (Empire325 Marketing, The State of AI Agency Recommendations 2026)
Key findings
What AI actually cites when a buyer asks it for an agency
Finding 01
The recommendation set is concentrated — and each engine has a different favorite
A short list of sources supplies most “best agency” recommendations, but which sources depends heavily on the engine. In Claude (June 12), firstpagesage.com appears in 5 of 16 verticals and thriveagency.com in 4 of 16 — across 124 unique domains. In Perplexity (June 13), the single most-cited source is reddit.com at 11 of 16 verticals, a heavy lean on community discussion; firstpagesage.com, g-co.agency, directiveconsulting.com, and thriveagency.com each appear in 4. In ChatGPT (June 13), dataally.ai leads at 5 of 16. There is no single “win the AI recommendation” target: the source that anchors Claude is not what anchors Perplexity or ChatGPT.
| Engine | Date | Top-cited source | Verticals | Type |
|---|---|---|---|---|
| Claude (claude-opus-4-8) | 2026-06-12 | firstpagesage.com | 5 of 16 | Listicle / roundup |
| Claude (claude-opus-4-8) | 2026-06-12 | thriveagency.com | 4 of 16 | Agency-owned |
| ChatGPT | 2026-06-13 | dataally.ai | 5 of 16 | Listicle / roundup |
| Perplexity | 2026-06-13 | reddit.com | 11 of 16 | Community / forum |
| Perplexity | 2026-06-13 | g-co.agency | 4 of 16 | Agency-owned |
Source: citations-claude-2026-06-12, citations-chatgpt-2026-06-13, citations-perplexity-2026-06-13 (raw counts of distinct verticals per domain).
Finding 02
Buyers are handed listicles, not agency homepages
In the Claude baseline the citation-type mix was 74.7% listicle/comparison (115 of 154 classified cited pages), 24.0% vendor/other (37), and 1.3% directory/review (2). Three out of four times an engine backs a “best agency” answer, it points at a “Top 15 [vertical] agencies” article — often one an agency wrote about its own category. The live July 22 SERP corroborates this: page-one results are dominated by self-published agency roundups plus a thin directory layer (Clutch, DesignRush, Digital Agency Network). About 11 domains recur in both the June in-model tracker and the July live SERP — genuine cross-method corroboration that this roundup-and-directory layer, not homepages, is the surface both AI and search draw from.
115 of 154 classified cited pages
37 of 154 classified cited pages
2 of 154 classified cited pages
Domains cited in both the June AI tracker and the July 22 live SERP:
nogood.iodirectiveconsulting.comamraandelma.comcollateral.comsilverbackstrategies.commintcopywritingstudios.comamsive.combeomniscient.cominfluencermarketinghub.comclutch.cog-co.agencyFinding 03
In-model citation and SERP ranking are different games
The most instructive case in the dataset is asset management. ChatGPT cited empire325marketing.com/industries/asset-management in-model on all three tracked dates (June 13, 15, 22), listing “Empire325” among the recommended agencies. Yet in the live July 22 organic SERP for the same query, Empire325 did not appear — that result set was led by Amra & Elma, ProperExpression, Bird, Arro Communications, and Select Advisors. Being cited in an AI answer and ranking on Google page one are correlated but not the same; a page can be indexed and AI-citable while not ranking organically.
An honest correction to our own headline number. The widely quotable “0/16” is Claude-only. The accurate framing is: not cited by Claude or Perplexity on any of 16 buyer queries; cited by ChatGPT for 1 of 16 verticals (asset management), consistently across three dates — not a blanket zero. We publish the correction because a study with our name on it is only worth the honesty of its smallest number.
Finding 04
The per-vertical view — there is no universal “best agency”
Live organic-SERP leaders for the 9 verticals we ran on July 22 differ almost entirely by industry. Only a handful of names — Amra & Elma, Directive, Siege Media, Mint Studios — bridge multiple verticals, exactly the multi-vertical listicle presence Finding 02 describes. NoGood (nogood.io) was the most pervasive single agency in the snapshot, a result link in 4 of the 9 queries.
| Vertical | Live-SERP leaders (2026-07-22) |
|---|---|
| Hedge fund | Amra & Elma, Collateral Partners, Fundmarketers, Croton Content |
| Asset management | Amra & Elma, ProperExpression, Bird, Arro Communications, Select Advisors |
| Financial services | Cramer-Krasselt, Amsive, Fintel Connect, Mint Studios |
| Healthcare | G&CO, Healthcare Success, Intrepy, Weinbach |
| B2B | Siege Media, Digital Agency Network, Altitude, Directive |
| SaaS | Animalz, Directive, The B2B Playbook, Rubicon |
| Fintech | Siege Media, Martal, Mint Studios, RightLeft |
| AI | Field Trip, Influencer Marketing Hub, Omniscient Digital, Tofu |
| Ecommerce | Clutch, DesignRush, Sweat Pants Agency, Silverback |
Single US-only WebSearch organic snapshot, 2026-07-22, no variance measure. Private equity, legal, insurance, education, manufacturing, biotech, real estate, and enterprise were not run live.
What it means
Four grounded moves to become the agency AI names
The battleground is the roundup layer, not your homepage
74.7% of Claude's supporting citations were comparison articles. Presence in credible “best [vertical] agencies” content is the actual entry mechanism — worth more than another homepage redesign.
Optimize per engine — the sources differ structurally
Reddit presence drives Perplexity, firstpagesage anchors Claude, dataally anchors ChatGPT. There is no single surface to win; each engine reads a different corner of the web.
Measure in-model, not by SERP proxy
The asset-management case shows AI citation and Google rank move independently — in both directions. Track the model answers themselves, not just rankings.
Vertical specificity is the replicable pattern
The one place a challenger broke through (ChatGPT, asset management, three-for-three) was its most vertically-dedicated industry page. Pick your highest-value verticals and win their citation sources one at a time.
This is the discipline we call Generative Engine Optimization — engineering your presence in the exact sources AI cites. See how it maps to a vertical in our industry agency guides and the earlier 2026 AI Recommendation Report.
Methodology & honest limitations
Engines and dates. Between June 12 and July 22, 2026 we ran the query pattern “best [vertical] marketing agency” across 16 verticals against three AI answer engines and one live search snapshot. (1) Claude (claude-opus-4-8): a single 16-query baseline on 2026-06-12 with daily re-runs through 06-25; we recorded per-query cited domains and whether any empire325marketing.com URL appeared. (2) ChatGPT: 16 queries on each of 2026-06-13, 06-15, and 06-22 (48 rows), recording both named brands and cited URLs. (3) Perplexity: 16 queries on the same three dates (48 rows). (4) A live US-only WebSearch organic-SERP snapshot on 2026-07-22 for 9 of the 16 verticals.
Counting rule. The query set is N=16 buyer queries, one “best [vertical] marketing agency” per vertical. A domain is counted “cited” for a vertical if it appears in that answer's cited URLs (named brands are tracked separately). All frequencies such as “5 of 16” are raw observed counts of verticals on the stated date — not percentages, not smoothed rates, not population estimates.
Limitations, stated plainly. This is not a full Perplexity/Gemini/Google-AI-Overviews panel — Gemini and AI Overviews are absent from the in-model data, and Perplexity/ChatGPT are three-date single-vendor samples. The live SERP snapshot is an organic-ranking signal, not an in-model citation; it was run once, has no variance measure, and covers only 9 of 16 verticals (private equity, legal, insurance, education, manufacturing, biotech, real estate, and enterprise were not run live), and organic SERPs drift between runs. We deliberately exclude the 51-day citation-probes.jsonl aggregate (792 rows, 05-29 to 07-20) from citation-rate claims: all 49 Anthropic direct-API probes returned HTTP 401 and only 253 of 792 rows are error-free (all OpenRouter/DeepSeek), with its Empire325-cited rows dominated by brand-name self-queries. Sample sizes are small (16 queries × 1–3 dates per engine); treat every figure as an observed data point.
Dataset: state-of-ai-agency-recommendations-2026.json / CSV — per-engine top domains, Claude's citation-type mix, cross-method recurring domains, and Empire325's per-engine status.
Author: Milton James Acosta III, Founder & CEO, Empire325 Marketing. Published 2026-07-22. Last reviewed 2026-07.
License: Creative Commons Attribution 4.0 (CC BY 4.0). Free to cite, quote, and republish with attribution to Empire325 Marketing and a link to this page.
Apply the research
Want to know which sources AI cites in your category?
We'll run this exact cross-engine analysis for your vertical and competitors, then map the specific sources you need to win. Book a 15-minute call and we'll share a citation-gap audit for your brand.
Get your citation-gap auditMore original research in the Empire325 Research library.