AI Visibility for Marketing Agencies: How AI Builds Your Prospects' Shortlists

Buyers rarely type "marketing agency" into an assistant. They describe their situation and ask for names. That makes the answer a shortlist, and every detail in the prompt is a filter your agency either passes or fails.

Here are realistic examples of what a CMO or founder might ask:

1. "Recommend B2B SaaS marketing agencies that work with Series A startups."
2. "Best paid social agency in Manchester for ecommerce brands under a $10k monthly budget."
3. "Which agencies specialize in healthcare SEO and understand compliance?"
4. "Compare [Agency A] and [Agency B] for a HubSpot migration."
5. "Is [Agency] still a certified partner for [ad platform]?"
6. "What clients has [Agency] worked with in fintech?"
7. "Should a 20-person company hire a fractional CMO or a full-service agency?"

Prompts 1 to 3 are discovery: you need to be named. Prompts 4 to 6 are verification: the prospect already knows you and wants the assistant to confirm a fact. Prompt 7 is category-level, where the answer may name no agency at all.

Write your own list of 10 to 20 prompts in these three groups before you measure anything. Keep in mind that engines rewrite prompts into their own search queries. OpenAI's help documentation says ChatGPT search "typically rewrites your query into one or more targeted queries," so your list is a sample of buyer intent, not the exact input.

Agency buyers use several assistants, and each one draws on different data. As of October 2026, treat them as separate channels:

• ChatGPT answers from its trained knowledge unless it decides to search. OpenAI documents a knowledge cutoff of Feb 16, 2026 for the GPT-5.6 family and Aug 31, 2025 for the Instant path, so a rebrand or new office after those dates is invisible to answers given without search.
• Google AI Overviews and AI Mode only cite pages that are indexed and eligible to show with a snippet, according to Google's AI features documentation. Google says they "may use different models and techniques," so check both.
• Perplexity and Claude search the web for many queries, using their own crawlers.
• Microsoft Copilot grounds its answers in the Bing index, which is separate from Google's.

Retrieval is discretionary on every platform. The same prompt can produce an answer with live search one day and from memory the next. So a single blended score hides the real picture. Measure each engine on its own, and note whether the answer cited sources. If your team is deciding who should run this work, see whether your current SEO agency should already handle AI search.

AI platforms get specific facts wrong about specific brands. Agencies are exposed because their facts change faster than most businesses' facts. These are the common failure points to check:

• Services. An agency that dropped web design two years ago may still be recommended for it, while its new specialty is missing.
• Specialties and niches. A generalist description buries a vertical focus, or an old niche sticks after a repositioning.
• Locations. Closed offices, a move to remote-first, or a city listed only on one directory page.
• Client claims. Former clients presented as current, logos the assistant cannot verify, or one client's work credited to a competitor.
• Partner status and certifications. Lapsed or upgraded partner tiers that directories still show in their old state.
• Pricing and minimums. A retainer minimum quoted from an old listing.
• Ownership and names. Mergers, acquisitions and rebrands, where the assistant merges two agencies or keeps the old name.

Say you run a hypothetical performance agency in Denver that moved to a remote team and stopped offering SEO. An assistant could still list your old downtown address and recommend you for SEO. The fix starts on your own site: state each current fact once, plainly, in the HTML of a page that can be fetched. A specific sentence ("We are a remote team serving US ecommerce brands, with a $5,000 monthly minimum") gives an engine something it can quote. Then make sure your service, about and contact pages agree with each other.

Superlatives do not substitute for facts. Read whether calling yourself 'the best' makes AI recommend you before rewriting your homepage.

When an engine answers a shortlist prompt, it often cites pages about agencies rather than agency websites. Being named on those pages is associated with appearing in the answer. Nobody can promise the effect, but a wrong fact on a heavily cited directory will keep resurfacing until it is corrected.

Check these, in this order:

1. Agency directories and review platforms. Services, locations, minimum budget, team size and client list. Update every profile you control and close duplicates.
2. Partner directories run by software and ad platforms. Partner tier, certified services and regions.
3. Award listings and shortlists. Correct agency name, the right office and the right client attributed to the work.
4. Trade press and "top agencies" roundups. You cannot edit these, but you can ask the publisher to correct factual errors.
5. Your team's and clients' public profiles. Current employer names and case-study references.

The fastest way to find which pages matter is to look at the sources the engines cite for your discovery prompts. Fix what you own, request corrections for what you do not, and record the date of each change. For how review volume fits in, see whether online reviews affect AI recommendations. If you also work in communications, the PR agencies guide covers earned media in more depth.

Before any of the above can work, AI crawlers have to reach your pages. These are the checks with the strongest evidence behind them.

Per-bot robots.txt rules

OpenAI documents that GPTBot governs training use and OAI-SearchBot governs ChatGPT search, and that "each setting is independent of the others." Blocking GPTBot to protect client work does not remove you from ChatGPT search. Blocking OAI-SearchBot does. Check OAI-SearchBot, PerplexityBot and Claude-SearchBot by name:

User-agent: OAI-SearchBot
Allow: /

Firewall and CDN blocks

A firewall or CDN can return 403 to a crawler your robots.txt allows. OpenAI's help documentation says eligibility needs both the robots.txt permission and a host that accepts traffic from its published IP ranges. Cloudflare changed its defaults on 2026-09-15, blocking Training and Agent crawlers on ad-displaying pages for new domains and existing free-tier customers. Check your settings even if nobody edited them. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.

JavaScript-only content

Agency sites lean on animated portfolios and case studies that load after JavaScript runs. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them." If your services list or client names only appear after hydration, those crawlers may not see them. Google renders JavaScript, so this gap mainly affects the other engines.

Snippet and index controls

A stray noindex, nosnippet, max-snippet:0 or data-nosnippet inherited from a template keeps content out of Google and Bing AI answers. Search your templates for them.

Agencies are often pitched tools that publish hundreds of AI-written posts to "own" AI search. OtterlyAI tested this in 2026 with two new domains carrying roughly 1,000 AI-generated posts each. Google deindexed one site after about 14 days and the other after about 90, and Otterly reports zero manual actions on either.

Google's spam policies describe this pattern directly, listing "Using generative AI tools or other similar tools to generate many pages without adding value for users" as scaled content abuse. Once the pages left Google's index, Google's AI surfaces stopped citing them. Microsoft Copilot, which grounds on Bing, kept citing one site: Otterly tracked 2,718 of its 3,934 citations from Copilot. That is one engine, on a low-stakes capybara topic, and it did not replace Google.

The test had one site per arm and no human-written control, so it cannot prove more than that. It is still clear enough for an agency: silent removal from Google is a bad trade for your own site and worse for a client's. More on this in whether AI content automation improves AI visibility.

Week 1: baseline

Run your prompt list on each engine separately. Record whether you are named, what each answer says about you, and which sources it cites. The free checker on this site gives a quick snapshot for up to three buyer questions.

Week 2: access

Check robots.txt per bot, firewall and CDN responses, JavaScript-only content, and noindex or nosnippet tags. Confirm index status in Google Search Console and Bing Webmaster Tools separately.

Week 3: your own facts

Write one plain statement for each current fact: services, niches, locations, partner status, minimum budget and current client categories. Put it in the server-rendered HTML. Remove contradictions between pages.

Week 4: third-party pages

Update directory, partner and award profiles. Request corrections from publishers. Log each change with a date.

Re-run the same prompts after the month. Expect slow movement: answers from trained knowledge only change when models are retrained. If you want to offer this to clients, the agency client reporting playbook covers how to report it honestly.