AI Visibility Index: U.S. Local-Service Marketing Agencies

Small Agency Edition — Fall 2026

A 20-agency, 30-response snapshot of which small U.S. local-service marketing agencies appear when buyers ask an AI assistant for a recommendation — and which never come up at all.

Marketing agencies help local-service companies get discovered. But are those agencies themselves visible when a business owner asks an AI assistant which marketing partner to hire?

SeenFirst tested ten unbranded, buyer-style questions across ChatGPT, Gemini, and Perplexity. We used fresh sessions, named no agencies in the prompts, and recorded 30 responses on September 14, 2026.

Before running any of them, we assembled a research cohort: 20 small U.S. agencies whose own public descriptions say they serve home-service, contractor and local-service businesses — precisely the firms these questions are asking for. The assistants were never shown that list, and no prompt ever mentioned a company by name. We then read all 30 answers and counted how many of the 20 the models named on their own.

The central finding was stark: only 2 of the 20 relevant small agencies in our research cohort appeared at all. Eighteen—90% of the cohort—were never mentioned in this 30-response snapshot.

That is not a judgment about agency quality. It is evidence of an AI-discovery gap.

The results

Rival Digital was the clear visibility leader within this cohort. It appeared five times, reached all three engines, and was placed first or second every time it appeared. Contractor Marketing Pros appeared once, in Perplexity, at position three.

Three findings worth watching

1. Relevance did not guarantee discovery

All 20 agencies were deliberately selected because their public descriptions showed clear relevance to home-service, contractor, service-area, or general local-service businesses. Yet only 2 of 20 surfaced in any tested response.

The practical lesson is narrow but important: a business can be a legitimate fit for a buyer's request and still be absent from the AI-generated consideration set.

2. Cross-engine visibility was rare

Only one cohort agency—Rival Digital—appeared in ChatGPT, Gemini, and Perplexity. Contractor Marketing Pros appeared only in Perplexity. The other 18 reached none of the three engines in this snapshot.

This matters because “visible in AI” is not one universal state. Different assistants can rely on different sources, retrieval systems, and response patterns.

3. The engines repeatedly favored a separate group of established names

Agencies outside the frozen cohort also appeared in these responses, several of them more than once. Because those companies were not part of the frozen cohort, they were not added retroactively and were not scored in the leaderboard.

We are not publishing response-level counts for them in this edition. Those counts were preliminary, and every figure in this article is one we can reconcile to the underlying response records — a standard we would rather hold than relax. A future edition will report non-cohort visibility once it is backed by the same evidence ledger the cohort results are.

Their visibility is nevertheless useful evidence: AI assistants had recognizable alternatives available even while most of the selected small-agency cohort remained absent.

A word about the eighteen

Every agency in this cohort was selected because its own public description showed clear relevance to home-service, contractor and local-service businesses. These are not marginal firms padded in to make a number look dramatic — they are specialists in exactly the work the test questions asked about, spread across 15 states and the full 1–50 employee range.

A separate technical review of all twenty websites found no failing site among them. Every one was reachable, secure, crawlable and indexable, and not a single broken piece of structured data turned up anywhere in the cohort. Several of the eighteen publish more complete machine-readable business information — address, telephone, service area, linked profiles — than either agency that did appear.

The gap this index measures is one of discovery, not competence. An agency can do the work, describe it clearly, build a technically sound site, and still not be part of the set an assistant returns. All twenty firms are named and linked in the cohort list below.

What we tested

• 20 U.S. agencies with 1–50 LinkedIn-listed employees
• 10 identical, unbranded buyer-intent prompts
• ChatGPT, Gemini, and Perplexity
• 30 total response records
• Fresh sessions and no follow-up questions
• Collection completed September 14, 2026

The cohort was purposively selected from our own prospect list of U.S. local-service marketing agencies — 110 ranked records, from which we drew twenty spanning 15 states and the full 1–50 employee range. We have commercial reasons to contact these firms, and had already approached some of them before publishing. We froze the list before collection, named no agency in any prompt, and publish every prompt, score and limitation here so the result can be checked independently. It is a research cohort, not an exhaustive list of every qualified U.S. agency.

How the score works

The SeenFirst Visibility Score combines observed mention rate, recommendation rate, top-three placement, cross-engine reach, official-domain citation, and identity accuracy. Every percentage is tied to the 30-response test rather than presented as an estimate of the wider market.

A score of zero means an agency was not observed in these responses. It does not mean the agency can never appear in AI results.

Inspect the method

Everything behind the numbers above is published here rather than summarized. If a figure in this article cannot be traced to one of these sections, treat that as an error worth reporting.

What this index concludes

Eighteen of twenty agencies that genuinely specialise in this work were absent from every answer. A separate technical review found nothing wrong with their websites, and several were better configured than the two that did appear. The assistants were not short of options — they named other firms readily, just not these.

So the finding is narrow and specific: in this snapshot, being qualified and being technically sound were not sufficient to be discoverable. Only one agency in twenty reached all three engines, which also means “visible in AI” is not one status a business either has or lacks — it varies by assistant.

We cannot say from thirty responses what does determine inclusion. We can say that most firms in this market appear to sit outside the set assistants return, and that no agency should assume it is inside that set without checking.

What an agency can do next

Start with a reproducible baseline. Identify which buyer questions surface the agency, which competitors appear instead, what sources the assistants cite, and whether the description of the business is accurate. Then improve the public evidence explaining the agency's identity, niche, services, proof, and relevance.

Generic “AI optimization” language is not a substitute for clear, corroborated evidence that machines and people can understand.

Technical appendix: was this just a website problem?

The obvious objection to these results is that the eighteen absent agencies simply have weaker websites. We checked.

On September 14, 2026 we ran a separate, read-only technical scan against all twenty cohort sites, looking at what a search or AI system encounters on the front page: whether the site responds, whether it can be crawled, whether it permits indexing, and whether it publishes machine-readable information about the business.

All 20 homepages returned HTTP 200 on the initial retrieval, were served over HTTPS, published readable robots.txt files, and exposed discoverable XML sitemaps. None of the 20 homepages carried a page-level noindex directive, and every homepage was retrievable on the initial scan.

Two sites — GoBIG Systems and LeadsForward — returned HTTP 403 to a second automated request made about a second after the first succeeded. We record that as a limitation of our own retrieval method rather than a finding about those sites.

Nor did machine-readable identity separate the two groups. Structured data describing the business as a local business was present on exactly half of the agencies that appeared, and exactly half of those that did not. Four of the most thoroughly configured sites in the cohort — complete business schema carrying address, telephone, service area and linked social profiles — were never mentioned once.

The agency mentioned most often does not publish that markup at all: no local-business schema, no postal address and no telephone number in machine-readable form.

We are not able to say what does explain the gap. In this homepage-level scan, the technical factors we measured did not distinguish the two agencies that appeared from the eighteen that did not.

One configuration worth borrowing

Among the twenty sites, one had a deliberately unusual robots.txt: it permits the AI agents that fetch a page at the moment someone asks a question, while declining the crawlers that collect training data. Traditional search crawlers are permitted too, and each group is labelled in the file with its intent.

That distinction matters more than it is usually given credit for. Blocking a training crawler is not the same thing as being invisible to an assistant — those are different systems, often operating under different agent names, and advice that treats “AI bots” as a single category misses it. The most-mentioned agency in this cohort was the one running that configuration.

One site is not evidence of cause, and we are not claiming it is. A robots.txt file also expresses a site's stated crawler preferences; it is not proof of how any system actually retrieved, or declined to retrieve, that page.

The paid SeenFirst audit currently tests ChatGPT, Perplexity, and Claude. This research index was collected using ChatGPT, Gemini, and Perplexity — that engine set describes the methodology for this study, not the product's current engine mix.