Why isn't our agency showing up in ChatGPT answers?

Because large language models cite brands that already have an authority footprint - so your agency stays invisible to ChatGPT until enough trade press, third-party experts and named reviewers have described what you do in language the model can lift.

That is the short version. The long version is below, with the three failures to check before commissioning new content and the cadence that compounds.

How the citation actually happens

When a user asks ChatGPT "what are the best agencies for B2B thought leadership in London", the model does not browse the web in real time and reason from scratch. It runs through the patterns it has already absorbed - large blocks of training data plus, increasingly, retrieved web context - and lifts the names that come up alongside the question pattern.

For your agency to come up, three things have to be true:

1. The web already associates your agency name with the question. Trade-press articles that name you, conference programmes that list you, podcast episodes that interview your founder, comparison pieces that include you.
2. The associations use language a model would lift. Concrete service descriptions, named verticals, named clients where permitted, specific outcomes. Vague descriptions - "award-winning agency", "creative partner" - get filtered out as boilerplate.
3. The associations are recent enough. AI engines in 2026 favour recency. A trade-press citation from 2024 is worth a fraction of one from this quarter.

If any of those is missing, the model has nothing useful to cite, and you stay invisible.

Why most agencies fail at this without realising

Most agency marketing has been optimised for the last decade to compete on Google's first page. The signals Google rewards - backlinks, on-page SEO, well-structured headings - are real, but they are no longer enough on their own.

LLMs reward a different shape: citation density. The number of distinct, authoritative third parties who have used your agency name in a sentence that an AI can usefully extract. You can rank number one on Google for a query that ChatGPT will still answer without naming you.

The mismatch is what most agencies are seeing this year. SEO numbers look healthy. AI citations do not exist.

The three failures to check first

Before you commission new content, audit what is already there.

Your name is not consistent. Your homepage says "comms agency". Your About page says "PR consultancy". Your LinkedIn says "integrated marketing communications partner". The model has no canonical hook to anchor your entity around. Pick one descriptor, write it identically on the homepage hero, the About page, the LinkedIn header, your podcast bio and your email sender name. Then keep it.

Your trade-press footprint is thin. Agencies that show up in AI answers usually have a steady cadence of byline articles in Campaign, PRWeek, The Drum, Marketing Week, Adweek. Not one piece a year. Three or four per quarter, with a recognisable named author. Volume threshold matters more than people expect; recent data from Brandi AI suggests brands producing twelve or more new or optimised pieces per month achieve up to two hundred times faster visibility gains than those producing four.

Your founder has no third-party-quoted opinion. If the founder has never been quoted in a publication that an LLM crawled, the model has no reason to cite them. Functionally, to the model, they do not exist as a thinker.

What to do next, in the order that actually compounds

1. Audit the entity. One descriptor, applied everywhere, no exceptions. Half a day.
2. Choose twenty to fifty buyer queries your real prospects would type into ChatGPT. Run them monthly. Track mention rate, position, and which competitors are named instead. The agencies winning this year publish the prompts they care about and grade themselves against them.
3. Build a trade-press programme before you build new content for your own site. Bylines in third-party publications carry far more AI weight per word than blog posts on your own domain.
4. Earn an author contributor URL. Once your founder has a permanent contributor page on Campaign, PRWeek or similar, that URL becomes a high-authority sameAs candidate for Person schema - and a stable address that LLMs return to.

How to measure whether it is working

A monthly run is enough.

Pick twenty queries that a buyer would realistically type. Mix category queries ("best B2B PR agency London"), comparison queries ("Hard Numbers vs Battenhall"), and use-case queries ("agency that helps with Cannes submissions"). Run each through ChatGPT, Claude, Perplexity and Gemini. For each engine, log: is your agency mentioned, what position in any list returned, what other agencies were named, what description was used.

The first three months are usually slow. By month four, one or two queries start returning your name. By month six, the queries you cared about begin to consolidate. The data drifts in the direction of whatever third-party material was published in the previous quarter, which is the strongest signal you will get that the trade-press programme is working - or that it is not.

Keep the same twenty queries for at least two consecutive quarters before refreshing the list. Otherwise you are measuring movement in the prompt, not movement in your visibility.

The real timeline

This work compounds. The first new visibility usually shows up in AI answers two to three months after consistent third-party publication begins. The shape after six months is dramatically different from the shape at month one. Agencies that demand visibility inside thirty days end up paying for vendors who promise it and do not deliver it.

The mechanic is not a shortcut. It is the same trade-press, expert-positioning, named-opinion engine that earns coverage in the old sense. AI search just rewards it harder.