Meikai × Le Figaro Media

Independent AI visibility measurement for branded content.

Meikai is partnering with Le Figaro Media, the advertising sales house of Le Figaro. Le Figaro Media will create branded content across its digital properties. Meikai will measure how brand representation and citations change in a defined set of generative AI responses.

What matters more than the partnership itself is that we are publishing the measurement protocol before the campaign runs, so the result can be judged against a standard set in advance, not one chosen afterwards.

Why earned media is worth measuring separately

A January 2026 University of Toronto study, Navigating the Shift: A Comparative Analysis of Web Search and Generative AI Response Generation, classified the sources four generative systems drew on. Earned media accounted for 46-65% of sources in the aggregate results and 59-86% for consideration-stage queries.

The study used a consumer-electronics query set, so those percentages are not universal benchmarks and should not be quoted as though they were. They do establish that editorial sources are a large enough share of what these systems retrieve to deserve measurement of their own, not a line in a report that only looks at owned content.

Why Le Figaro

Founded on 15 January 1826, Le Figaro is France's oldest national newspaper still in circulation. An established newsroom with that reach is a reasonable place to test whether authoritative branded content is retrieved or cited in AI-generated answers.

It is a test, not an assumption. Domain authority does not automatically transfer to AI visibility, model behaviour varies by prompt, platform and time, and the honest position before the campaign is that we do not know how much of an effect to expect. That is why the protocol is repeated measurement, not a one-off read.

The measurement protocol

Fixed in advance, so it cannot be tuned to the result:

• Prompt set: 50 defined prompts, held stable for the duration.
• Cadence: run daily, not sampled at the end.
• Platforms: more than six AI experiences, including ChatGPT, Perplexity and Gemini.
• Captured per response: whether the brand appears, how it is described, which domains and pages are cited, sentiment, and competitive share of voice.
• Comparison: a pre-campaign baseline, with results read at response level, not as a single index.

An observed change against that baseline establishes an association with the campaign period. It is not proof of causation, which would need a control or an experimental design, and we will say so when we report, however the numbers land.

The same protocol works for a newsroom

The protocol above is pointed at a brand, but a publisher can run it on itself, and the question is a live one for most newsrooms: are AI assistants citing our reporting, or are they citing someone who covered the same story. Nothing in the method requires a data team. It requires a fixed set of questions, a daily run and a record of which domains are cited in each answer.

Reading it by topic, not by article, makes it usable editorially. Citation share per beat shows which desks are being treated as a source and which are being passed over, and comparing your domain with named rival titles on the same questions turns a raw count into a position. Report each engine separately, because a title can be cited routinely in one and be absent from another on the same story.

It is also worth running when nothing has changed on your side. Platforms revise how they retrieve and weight sources on their own schedule, and a syndication change or a template change can move attribution without anyone intending it. A standing measurement surfaces that, instead of leaving you to discover it a quarter later.

The measurement window has to outlast the campaign

One design decision is worth explaining, because it is the sort of thing that is usually discovered too late.

Between March and July 2026 we tracked what happened to citations after a publisher removed a directory of pages, briefly restored them in June, then removed them again. Measured as citations per 1,000 responses for a fixed cohort of 126 brands, the 15-21 July window put Perplexity at 109.5% of its March baseline, ChatGPT at 12.2% and Google AI Mode at zero. The full analysis sets out the method.

That establishes one thing precisely: platforms let go of content at very different speeds. It does not establish the reverse. We have not measured how quickly each platform picks new content up, and we are not going to assume the two are symmetrical because it would be convenient here.

It does make a short measurement window hard to justify. If retention after removal varies by a factor of nine across platforms, treating acquisition as instant and uniform is an assumption, not a default. So the window extends past the campaign, and each platform is reported separately instead of averaged.

SEO remains the foundation

Crawlable pages, clear authorship, accurate structured data and genuinely useful content still do the load-bearing work. Google's own guidance for AI features is explicit that the same foundational SEO practices apply, that there are no additional technical requirements, and that no special markup buys inclusion.

Editorial coverage contributes something different: independent context that can be retrieved, summarised or cited. The useful question is not whether a given publisher carries some universal citation weight, but whether its content shows up in the specific answers and prompts you are measuring.

GEO spans several teams

Media partnerships, earned coverage, on-site content, product pages, structured data and audience research all shape how a brand is represented, which means the measurement needs input from media, PR, digital marketing, e-commerce, data and insights.

Meikai gives those teams one view across models, prompts and markets. This partnership points it at a narrow, stated question: whether and how the campaign coincides with changes in brand representation. Our source influence framework covers the broader model.