Brand SERP: how to control the first impression AI sees

For more than a decade, a brand's first impression was decided in one place: the page of results that appears when someone types your name into Google. A prospect who googles you before a meeting, a journalist who verifies before quoting you, a candidate who looks you up before accepting the offer. That page —the brand SERP— was the most honest mirror of how the world sees you, and arranging it was for years one of the most profitable disciplines in digital marketing.

The trouble is that now there are two mirrors. The second one isn't controlled by any communications team: it's what ChatGPT, Perplexity or Google's panel answers when someone asks about you. And that second mirror is becoming, for a growing share of people, the first one they look at.

The good news, if there is any, is that both mirrors are fed from the same source. Understanding that connection is what separates the brands already working on their AI presence from the ones that will discover it once it's too late.

What a brand SERP is, and why it mattered so much

The term was coined by Jason Barnard, of the consultancy Kalicube, and his definition is deliberately simple: "what appears on search engines when someone searches your brand name". It isn't the generic category search —"SEO tools", "invoicing software"— but the search for your own name.

Barnard has an observation that aged well: at some point in the buying journey, almost everyone will look you up by name. The client who got your proposal, the investor evaluating the round, the partner considering an integration. When that page loads, he says, the perception it forms "could very well tip the balance of the decision they are making." An ideal brand SERP is three things at once: accurate, positive and convincing. And the convincing part, he adds, comes from visual richness: the knowledge panel on the right, the sitelinks, the video and image boxes that only appear when Google understands who you are as an entity.

That last point is the hinge for everything that follows. Because to show a knowledge panel, Google first has to understand you as an entity: a node with attributes and relationships, not a string of text. And it turns out that understanding entities is exactly what language models do too.

The second mirror: what AI decided to show

On January 30, 2026, OpenAI made a change that went relatively unnoticed outside technical circles. According to its own release notes, ChatGPT's answers became "more visual and easy to scan": important people, places, products and ideas are now highlighted within the text, and tapping one opens a side panel with a summary, key facts, images and links to sources it considers trustworthy.

If that sounds familiar, it's no accident. It is, in essence, a knowledge panel built inside a chat. The deeper difference is that nobody on your team verifies that card, nobody suggests edits, and there is no Google form to claim the entity. The panel is assembled from what the model learned about you during training and what it retrieves in real time. Your first impression, again, but this time on a surface that shows up in no dashboard.

An honest caveat: not every query triggers source retrieval. A large share of answers are generated directly from the model's memory, without looking anything up live. That means how AI describes you depends, to a large degree, on the footprint you left in its training data long before anyone asked. It isn't something you fix on Tuesday for Thursday.

What connects both mirrors: entity signals

This is where the evidence becomes useful. In 2026, Ahrefs published an analysis of 75,000 brands looking at which signals best correlate with visibility in Google's AI Overviews. The three strongest turned out to be entirely off-site: brand mentions across the open web topped the list with a correlation of 0.664 —the single highest predictor—, followed by branded anchors in links (0.527) and branded search volume (0.392).

That ranking has an uncomfortable reading for anyone who grew up optimizing titles and meta descriptions: the levers that move AI visibility most live outside your site. They aren't what you write on your page, but how much and how the rest of the internet names you. It is, almost point for point, the same raw material that builds a good brand SERP: consistent mentions, a recognizable entity, people who search you by name.

A detail from the same body of research: branded search volume also correlates with chatbot mentions (around 0.334), a figure considered decent in this field. Put another way, the more people search you by name, the more likely a model knows you well enough to name you without getting it wrong. The brand SERP and the AI answer aren't two separate projects; they're two outputs of the same entity work.

What makes a model cite you (and what doesn't)

When retrieval does happen, the mechanics of which text gets cited are starting to come into focus. The analyst Kevin Indig published in February 2026, through the Gauge platform, a study of 1.2 million ChatGPT responses that examined more than eighteen thousand cited passages. Three findings help orient content work.

First: entity density matters. Cited content showed around 20.6% entity density —proper nouns, brands, tools, people— against 5 to 8% for standard English. A text that names concrete things gets cited more than one that generalizes. Second: 44.2% of citations came from the first 30% of the page; what you bury at the bottom is barely cited. Third: question-format headings correlated with roughly twice the citations.

None of the three is magic, and it's worth saying so: correlation isn't causation, and the study itself acknowledges it doesn't separate the retrieval, attention and generation processes. But the direction is consistent with everything else. Be specific, name real things, put the important part up top, structure with questions. That's good writing, not an optimization trick.

The cost of not looking at the second mirror

There's a concrete reason this is more urgent than it seems: the models get it wrong, and they get it wrong often. Columbia's Tow Center for Digital Journalism documented that the citations AI search engines generate are incorrect more than 60% of the time. If the model builds your card from wrong or outdated sources, there is no "report an error" button like in Google's knowledge panel. The fix runs, again, through repairing the sources the model reads.

This inverts the classic priority. For years, arranging the brand SERP was a "brand hygiene" task you did once and revisited now and then. Today it's a living system: the same entity signals that arrange your human first impression determine your algorithmic first impression, and the latter updates on cycles you don't control. Looking away doesn't freeze it; it leaves it adrift.

What we think makes sense to do

An opinion, to be taken as such. The first thing is the most boring and the most profitable: consistency. Make your site, your social profiles, your company listing and any directory you appear in all say the same thing about who you are and what you do. Contradictions between sources are what confuse both the knowledge panel and the model. One canonical description, repeated without odd variations, is worth more than ten clever versions.

Second: mark up your organization with structured data. An Organization schema block with the sameAs field pointing to your official profiles gives Google —and by extension the entity ecosystem— an explicit declaration of who you are and which accounts you identify with. It isn't optional if you want to be understood as an entity rather than a string of text.

Third, and slowest: brand mentions across the open web. The Ahrefs study was clear that this is the strongest predictor, and it isn't bought overnight. It's built by showing up on sites where your audience already is, earning reasonable press, being cited by others because you have something to say. It's foundational work, not a campaign.

And fourth, which almost nobody does: measure the second mirror. Google your brand with a critical eye once a quarter and ask ChatGPT and Perplexity who you are, as if you were a skeptical prospect. Write down what they get wrong. That list of errors is, almost always, a list of sources to repair.

Questions we get a lot