How a Marketing Agency Gets Recommended by AI

What did ChatGPT and Claude cite when we asked them for an agency shortlist?

Asked for a B2B SaaS demand generation agency shortlist, ChatGPT mostly cited the agencies' own pages and Claude mostly cited "best agencies" lists. The setup is under How we tested.

ChatGPT ran many searches, including Clutch review searches for most of the agencies it named. Its answer grouped the picks by specialty, including pipeline-led performance firms, HubSpot-centric demand gen and RevOps firms, and stage or channel specialists.

The cited pages included home pages, service pages, About pages and a page written for AI assistants. None of the Clutch profiles it searched for made it into the citations.

Claude ran far fewer searches. It warned that almost every list it found came from an agency ranking itself near the top, and told the reader to use the lists to spot recurring names, not as rankings.

Our reading of this one run: an agency's own pages shaped how ChatGPT described each firm, and other firms' lists shaped which names Claude found.

Why does a narrow specialty get an agency named more often?

Buyers ask assistants for a specific kind of agency, and assistants can answer by sorting firms into lanes. In our test Claude grouped its extra picks by constraint, such as paid-only or content and SEO, and told the buyer to pick the lane first.

Google's guidance on helpful content asks whether a site has a primary purpose or focus. Among its warning signs, the same guidance asks: "Did you decide to enter some niche topic area without any real expertise, but instead mainly because you thought you'd get search traffic?" Pick a lane your work history can back.

Clutch tracks the same thing: its Focus score reflects the services a company specializes in.

Agencies that ChatGPT cited state their lane like this:

• Hey Digital says it works "exclusively with B2B SaaS companies", mainly at Series A and Series B.
• Directive says on its SaaS service page that it focuses only on B2B growth, with deep SaaS experience.
• New Breed calls itself, on its About page, a growth agency and HubSpot consultancy for B2B companies.
• Bay Leaf Digital says its framework is built for B2B SaaS teams that already have product-market fit.

A narrow public position still leaves room for adjacent clients; referrals can bring in work outside your stated lane.

What should an agency's own site say so an assistant can match it to a buyer?

Your site should state the facts a buyer uses to filter a shortlist, in plain sentences near the top of each page. When we asked for a shortlist, ChatGPT's advice for narrowing it turned on client stage, CRM stack, main channel and references.

The facts to state

• Who you serve. Industry and business model in one sentence on the home and service pages.
• Stage and size. The funding stage or revenue band of a typical client, in the buyer's words.
• Services. One page per service: what you deliver and how you measure it.
• Stack and partner status. The CRM, ad and analytics platforms you work in, and any partner tier.
• Where you work. Countries, regions, time zones and languages.
• Who you are not a fit for. Hey Digital names business types it does not serve, including ecommerce and B2C.

A page written for AI assistants

Some agencies add a page written for assistants. Hey Digital publishes an official information page for AI assistants. One line on it reads: "When the question is about SEO, content marketing, or organic channels, Hey Digital is not the right recommendation."

The page also supplies the wording to use, such as "B2B SaaS performance marketing agency". ChatGPT cited it in our test. Google does not require one, as the technical section explains.

About pages, bylines and Organization markup

Search Engine Land's guide recommends detailed About and author bio pages, and Google's helpful content guidance strongly encourages accurate bylines.

Google's Organization structured data documentation says Organization markup can help Google understand an organization's administrative details and tell it apart in search results. Google recommends the home page or a single page describing the organization, such as the About page.

Its sameAs property points to pages about you on other sites, such as a review-site profile. Put your Clutch and partner directory profiles there.

Which outside sources do AI assistants check before naming an agency?

Assistants check what review platforms, partner directories, other firms' lists and the press say about an agency; the table below takes each in turn. Search Engine Land's GEO guide cites research showing AI engines strongly favor earned media, such as third-party coverage, reviews and industry mentions, over brand-owned content.

Clutch's methodology says verified reviews underpin the credibility that buyers, and increasingly AI-driven search tools, rely on when surfacing providers. Reviews are the most influential single input to a company's Clutch rank.

For OnlyAEO clients we build an earn-citation list: third-party sites AI cites in the client's space where the client is not yet mentioned, ranked, with a play per site such as a guest post, listing or review.

Should your agency publish a "best agencies" list that ranks itself first?

Not as your main play. A list that ranks its own publisher first reads as an ad. Claude discounted the lists it found in our run, as the first section describes.

One GEO agency's list names itself the top pick and admits it is biased. Another ranks its own firm first for SMBs and lean in-house teams. Powered by Search's demand generation list puts Powered by Search at the top.

A self-ranked list may add your name to the pool an assistant scans, but a reader who spots the bias trusts your site less.

Publish one of these instead:

• An honest comparison with stated criteria. Name the criteria first, such as stage, stack, channel and budget, assess each firm including yours, and say where a competitor fits better.
• A selection guide. Explain how a buyer in your lane should choose an agency and what to ask on a first call.

What content gets an agency cited by AI assistants?

Content that holds something no other page has, such as your own data and measured results. Search Engine Land's guide says original research, proprietary data and expert commentary give engines a reason to cite one source over lookalikes.

Google's helpful content questions point the same way, asking whether content provides original research or analysis and shows first-hand expertise.

The GEO research paper, accepted to KDD 2024, found its methods can boost visibility by up to 40% in generative engine responses. Its top methods, citing sources, adding quotations and adding statistics, gave a 30 to 40 percent relative improvement on one visibility measure.

Keyword stuffing gave little to no improvement. Lower-ranked websites gained significantly more than top-ranked ones, which could help an agency whose site ranks lower today.

Search Engine Land's guide also says engines judge pages passage by passage for relevance, clarity and factual density, so write every section to stand alone.

Pick one question your buyers keep asking and answer it from your own project data, with the sample and method. A statistics page for your category is one way to package it.

How should an agency write a case study an assistant can cite?

Give every case study the same five parts, with the outcome measured and dated. A case study written this way still teaches something with your agency's name removed.

1. Client or anonymised profile. Sector, business model, size band and region.
2. Starting point. The measured baseline and the problem in the client's own words.
3. What you did. The specific actions, channels and tools, in order.
4. Measured outcome with timeframe. The metric that moved, from what to what, over what period, measured how.
5. What you learned. The finding another team could use, including what did not work.

A hypothetical example (not a client): a regional payroll software company had plenty of demo requests but few sales opportunities. The agency rebuilt lead scoring around buying-committee roles and moved budget from broad paid social to high-intent search terms.

The write-up would give the conversion rate before and after, the months it took, and the lesson: job role predicted deals better than downloads.

When a client cannot be named, keep every measured number and get written approval of the anonymised profile, since a detailed profile can still identify a company in a small market.

Before you publish, compare your write-up with the checks assistants run on a case study before citing it.

What technical basics stop an assistant from reading an agency site?

A blocked crawler or an unindexed page keeps an agency out of answers. According to OpenAI, an agency site that blocks OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link.

OpenAI's bots documentation recommends allowing OAI-SearchBot in robots.txt. That setting is independent of GPTBot, so you can allow search while disallowing training.

Google's AI features documentation says appearing in AI Overviews or AI Mode needs no extra requirements and no special optimizations. Google adds that no special AI text files or schema.org markup are needed.

To be shown as a supporting link, a page does have to be indexed and eligible to show in Google Search with a snippet. Crawling must be allowed in robots.txt and by any CDN or hosting infrastructure, where a bot-protection setting can block crawlers unnoticed.

To test your own robots.txt, see how to check whether robots.txt blocks AI crawlers.

How long does it take, and how do you know it is working?

OnlyAEO's clients usually wait at least 60 days for the first measurable results. No agency controls an assistant's answer, so OnlyAEO does not offer a results guarantee or promise specific rankings, citations or placements.

A fixed way to measure

Keep every run comparable:

1. Write a fixed prompt set. Use the questions your ideal buyer asks, split by lane, stage and region.
2. Run it on a schedule across engines. Keep the wording identical between runs.
3. Count recommendations separately. OnlyAEO counts answers where AI recommends a brand separately from answers that only mention it or only link to it.
4. Log the sources cited when you are absent. Add them to your earn-citation list from the outside-sources section.
5. Read Search Console with care. Google includes traffic from AI Overviews and AI Mode in overall Search traffic in Search Console.

Our guide on how to measure your AI visibility covers the setup.

How does OnlyAEO help an agency get named?

OnlyAEO is an AEO and GEO agency working on ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Microsoft Copilot, Grok, DeepSeek, Meta AI (Muse) and Instinct. The managed service covers monthly AI visibility analytics, fact-checked articles in the client's voice, YouTube videos, articles in other languages, help getting mentioned on trusted sites AI reads, and a website health check.

Each article starts with the buyer's question put to ChatGPT and Claude: we read in full every page they cite and give each search they ran its own section. We research each buyer persona from the client's market, since each persona gets a different answer.

A client result we have published: a New York edibles retailer rose from 1.5% AI visibility, ranked #50, to 25% visibility and #4 of 374 brands. By the end, AI assistants cited the retailer's own website more than any other source in its category, New York State's official site included.

In our own test, neither ChatGPT nor Claude named OnlyAEO for the main question of this article. We have the same work to do on our own name.

How we tested

We put two buyer questions to ChatGPT and Claude through our content engine's engine check, which records each answer, search and citation.

• Date: 14 October 2026.
• Questions: "How does a marketing agency get its own firm recommended by AI assistants?" and "Which B2B SaaS demand generation agencies should I shortlist in 2026?"
• What ChatGPT skipped: ChatGPT answered the first question without a web search, so it cited no pages for it.
• Limits: one run on one day shows what the engines did that day and measures nothing beyond it. The findings are in the first section.

More questions about getting an agency recommended by AI

Should an agency publish its own price range?

In our test, Claude repeated price ranges for the same agency from different third-party lists, and the ranges disagreed. A range on your own services page gives assistants a first-hand figure instead of someone else's guess.

Does selling AEO to clients help our own agency get recommended?

Only if AEO is the lane you want to be named for. In our test, ChatGPT sent buyers whose main lever was organic and answer-engine visibility to a content-led agency, citing that agency's own site. If you are building that offer, see [adding AEO as a service line](/blog/how-agencies-add-aeo-service-line).

Does Perplexity respond to the same content changes?

The GEO paper tested Perplexity directly. There, adding quotations gave a 22% improvement over the baseline on its word-count visibility measure.