There is a new kind of pitch happening in marketing, and most agencies are not in the room.
A brand manager opens an AI assistant and types: "Find me a digital agency specialising in retail marketing in South Africa with experience in loyalty programmes." The AI searches, synthesises, and returns a shortlist. Some agencies are on it. Most are not. The ones that are not have not been disqualified by their work, they have been disqualified by their website architecture.
This is the invisible agency problem, and it is accelerating.
For decades, being discovered meant ranking on Google. Agencies hired SEO specialists, published thought leadership, optimised meta descriptions, and built backlinks. That game is not over, but a second game has started alongside it, one with different rules, different referees, and almost no agencies playing it deliberately yet.
The new referees are AI systems. And they read very differently from humans.
Who Is Actually Reading Your Website
When a brand's procurement team uses an AI tool to compile a vendor shortlist, that tool is not browsing your site the way a human would. It is not admiring your case study carousel or watching your showreel. It is attempting to extract structured facts: what do you do, for whom, with what results, at what scale, in which sectors.
If your homepage is a JavaScript-rendered single page application that returns a blank HTML shell until the browser executes several hundred kilobytes of code, the AI sees nothing. If your services are described as "end-to-end brand experiences" and "integrated storytelling solutions" rather than "social media management," "paid search," and "retail activation," the AI cannot categorise you. If your case studies live inside a beautifully designed PDF that requires a download and a login, the AI cannot read them at all.
The uncomfortable truth is that most agency websites, which have been optimised for human impression, are largely opaque to machine comprehension. And as more of the buying journey gets mediated by AI tools, from initial discovery through to briefing, evaluation, and even contract drafting, opacity has a direct cost.
It is not a theoretical future cost. It is happening now, in every RFP process where an AI assistant compiles the initial vendor list, in every startup founder who asks Claude or ChatGPT to recommend agencies for their launch, in every brand team that uses an AI procurement tool to benchmark capabilities before a chemistry meeting.
The AI Reputation Gap
Ask your agency name to ChatGPT. Ask it to Claude. Ask Perplexity. Notice what comes back.
If your agency is mentioned at all, the description is almost certainly constructed from whatever text is most consistently and clearly published about you on the public web. It will reflect your Wikipedia page if you have one, your Clutch or Agency Spotter profile, press releases, LinkedIn, and whatever fragments of your website were successfully crawled before JavaScript rendering blocked the rest.
That constructed description may be accurate. It may also be years out of date, missing your most significant capabilities, or conflating you with another agency of a similar name. And crucially, you probably have no idea what it says, because you have never systematically checked.
This is your AI reputation. It exists whether or not you have chosen to manage it. The agencies that will win the next decade of new business are the ones that decide to manage it deliberately, starting now, while the field is still open.
Five Things Agencies Must Do
1. Make the Website Actually Crawlable
The first and most consequential fix is structural. Agency websites are frequently built on JavaScript-heavy frameworks that prioritise visual experience over machine readability. This made sense when the only non-human reader was a Google crawler, and Google had largely learned to execute JavaScript. AI retrieval systems have not. Many LLM-based tools, RAG pipelines, and AI procurement assistants work from static HTML only.
The test is blunt: open your terminal and curl your homepage. Read what comes back. If the meaningful content, the services, the positioning, the proof points, is absent from that raw response, you are invisible to a large class of AI systems.
The fix does not require rebuilding from scratch. It requires ensuring your site is statically rendered or server-side rendered, so the content exists in the HTML before any JavaScript runs. Modern frameworks support this. Your development partner knows how to do it. The agency that actually prioritises it in the next six months will have a structural advantage over every competitor that does not.
2. Write for Machines as Well as Humans
Agency copywriting is designed to create impression and emotion. That is appropriate for human readers. It is useless for machine readers, which need specificity, consistency, and extractable facts.
This does not mean replacing your brand voice with technical boilerplate. It means layering in structured clarity beneath the creative expression. Every service page should contain an explicit, jargon-free description of what the service is and what it produces. Every case study should include a structured summary, client sector, brief, approach, measurable outcome, tools used, even if the main body of the case study is rich and narrative. Every team page should clearly identify roles and areas of expertise, not just names and titles.
Think of it as a second document living inside the first. The human reads the polished version. The machine reads the structured layer. Both can coexist on the same page.
Add Schema.org JSON-LD markup to your key pages. This is a standardised vocabulary for declaring what an entity is, what it does, and how it relates to other entities. An agency website with proper Schema.org markup is, to a machine, categorically easier to understand than one without it. Your web team can implement this in a day. It will pay dividends for years.
3. Publish an llms.txt File
The llms.txt convention is modelled on robots.txt,the file that tells search engine crawlers which parts of a site they may access. Where robots.txt sets access rules, llms.txt provides curated, plain-text guidance for language models: here is what we are, here is what we do, here are the most important pages on our site, here is how to reach us.
Placed at youragency.com/llms.txt, this file becomes the first thing an AI system that respects the convention will read before attempting to interpret anything else on your site. It gives you authorial control over your AI-facing narrative in a way that no other mechanism currently provides.
For an agency, an effective llms.txt would include a plain-language description of the agency's positioning and specialisation, a list of service areas with brief, specific descriptions, a pointer to the most important case studies, a note on geographic focus and client size, and contact routing for new business enquiries.
This is not a technical task. It is a writing task. It takes an afternoon and it immediately improves how every LLM that queries your domain constructs its understanding of you.
4. Restructure Your Case Studies
Case studies are the most important content an agency produces for new business purposes, and they are almost universally formatted in ways that make them machine-unreadable.
A case study that lives inside a PDF requires a download. A case study that exists only as an image requires optical character recognition to parse. A case study behind a login wall is invisible to any external system. A case study that leads with three paragraphs of scene-setting before mentioning the client's sector or the campaign's objective forces a machine to infer what a human would skim to find.
The solution is to publish case studies as indexed, static HTML pages with a consistent structured format. The header of every case study should declare the client sector, the brief in one sentence, the core solution in one sentence, and the primary measurable result. The body can then expand freely into narrative. This approach serves human and machine readers simultaneously, the structured summary satisfies the parser, the narrative satisfies the reader.
Over time, this structured case study library becomes a powerful signal to AI systems evaluating your agency's track record. When a machine asks "has this agency done loyalty programme campaigns for FMCG brands in emerging markets," it can answer that question from your case study metadata rather than having to read and interpret unstructured prose.
5. Audit and Manage Your AI Footprint Actively
Your AI reputation is a live, dynamic thing. It changes as new content about your agency is published and indexed, as the models are updated, and as the web evolves. Managing it is not a one-time task, it is an ongoing practice.
This means regularly querying the major AI systems about your agency and tracking what they return. It means identifying gaps, capabilities that are real but not reflected in AI outputs, and closing those gaps through published content. It means monitoring third-party profiles on Clutch, G2, Agency Spotter, and industry directories, because these are often the sources AI systems rely on most heavily when direct website content is unavailable.
It also means being thoughtful about where your agency is mentioned on well-indexed external sites. A well-written profile in a respected industry publication, a quoted comment in a trade press article, or a listed partnership with a technology vendor are all signals that AI systems use to construct their understanding of you. These are, in the AI era, the equivalent of the backlinks that shaped SEO rankings, and they deserve the same deliberate attention.
The Deeper Opportunity
Everything described above is defensive, it is about not being invisible. But there is an offensive dimension too, and it is larger.
Agencies that develop genuine AI fluency, not just literacy, but operational fluency, where AI is woven into how strategy is developed, how creative is produced, how reporting is built, and how client relationships are managed, will be able to deliver work that agencies without that fluency cannot. The gap between AI-fluent and AI-illiterate agencies is currently small and closing rapidly from one direction. The agencies investing in that fluency now, training their teams systematically rather than letting individual practitioners find their own way, will find the gap has become a moat by the time their competitors notice it.
The agencies most at risk are those in the middle, large enough to have legacy processes that resist change, not large enough to have the resources to transform quickly. The small, nimble agencies and the genuinely large ones with serious innovation budgets are both moving. The mid-tier agencies that assume their existing reputation will carry them are the ones that will find, two years from now, that they are being left off shortlists they have no idea they were never on.
Starting Tomorrow
The full transformation of an agency's AI presence is a six-to-twelve month project. But the first steps take days, not months.
Check whether your site renders without JavaScript. Query your agency name in three AI systems and read the results carefully. Write an llms.txt file. Add a structured summary block to your five most important case studies. Ask your web team what it would take to add Schema.org markup to your service pages.
None of this requires a strategy consultant or a digital transformation budget. It requires the same thing that good agency work has always required: the discipline to see the world through the eyes of your audience, and the craft to communicate clearly to them.
Your audience now includes machines. They are not harder to communicate with than humans. They are just different. And right now, almost no one is talking to them at all.
