Content design · Birdeye · 2024–2025

Designing the language layer for Search AI

Search AI is a Birdeye product that helps businesses track and improve their visibility in AI-generated answers from ChatGPT, Gemini, and Perplexity. I was the content designer across the full product, from dashboard microcopy to landing page to the content system that kept it consistent.

Lead Content Designer
Full product
3-month launch window
Employee of the Month

A new product category needed a new content vocabulary

Search AI sits at the intersection of local SEO, AI citations, and brand reputation. None of these had established UX language when the product launched. When someone asks ChatGPT "best dentist near me," your business either appears or it doesn't. Search AI helps you track that, understand why, and fix it.

The challenge wasn't just writing copy. It was building a language system from scratch, for a product category that didn't fully exist yet, under a 3-month launch deadline.

Dashboard copy: section titles, body text, labels, tooltips
Landing page: H1, H2s, value cards, FAQ, CTAs
Email copy: report-ready notifications, onboarding flows
Content system: Brand vs Location language rules
Microcopy: banners, alerts, error states, empty states
Stakeholder alignment: translating engineering logic to UX copy

Four decisions that shaped the product's voice

These weren't just copywriting calls. Each one required diagnosing a problem, making a case, and sometimes pushing back on what already existed.

1

Rewriting the visibility score banner

The PM's draft was technically accurate but would have alarmed users. Engineering logic needed to become user reassurance.

PM's draft

"Visibility Score Update: Now includes rank in LLM searches (was 100% just for appearing). Rank 1 = 100%, drops with lower ranks. Score fell? Algorithm change only, optimize for #1!"

Final copy

"Visibility score now reflects ranking in AI results. Higher positions contribute more to your score, so changes may be due to this update."

Why I changed it: "LLM searches" is internal jargon. "Score fell?" triggers panic before explaining why. The exclamation mark undercuts trust. I translated the engineering logic (a weighted ranking formula) into one calm, accurate sentence. I also gave my colleague the reasoning so they could defend it in the team group chat without me being in the room.

2

Pushing back on a generic H2

The landing page needed section headings that explained mechanism, not just outcome. Generic was easy. Specific took longer but earned it.

First draft

"See why AI recommends certain brands over others"

Final copy

"Decode what AI looks at before recommending your brand"

Why I pushed for this: The first version explains nothing new. It creates no tension. "Decode" signals insight and mechanism. It tells the reader this section will actually teach them something. "What AI looks at" explains the process, not just the outcome. I flagged the original as too generic and held the revision through multiple iterations until it earned its place.

3

Standardising dashboard section structure

The dashboard had an inconsistent mix of question-format titles and statement titles. A system decision was needed, not a one-off fix.

Before (inconsistent)

Section 1: "Overview"
Section 2: "How is Search AI score calculated?"
Section 3: "Business performance by location"
Section 4: "How are your themes & prompts performing"

After (pattern established)

Question title + supporting body text, applied consistently.

"How is your Search AI score calculated?"
See the key metrics that contribute to your overall Search AI score.

Why this mattered: Inconsistency across a dashboard erodes trust even when individual copy is good. I established a clear pattern, question title followed by body text that expands it, and applied it across every section. The table headers also got fixed: "Impact" became "Expected impact" and "Impact metric" became "Metric affected." Small changes, but they make the product feel designed rather than assembled.

4

Defending question-format titles without question marks

The CPO pushed back on dropping the question mark from section titles. I held the position, explained the reasoning, and he agreed. The pattern shipped as I proposed.

Figure: reference screenshot from the full case study (embedded image omitted in this web export).

Why I held this: These titles aren't real questions. They don't prompt an answer from the user. They frame a data view. A question mark creates a false expectation, implying the section will respond to something the user asked. Search AI is a genuinely new concept and users arrive with a lot of questions about AI visibility. Starting sections with "How" and "What" matches that mental state without turning the dashboard into a FAQ. NN/g and the Microsoft Style Guide both recommend minimising punctuation in UI headings for exactly this reason. Products like Amplitude and Mixpanel use the same Wh-word pattern in their dashboards without ending punctuation. Anil agreed, the pattern shipped, and no user confusion was reported.

Brand vs Location: a language rule that prevented drift

Search AI has two main views: performance by location and performance by brand. Without a clear language rule, copy would bleed across views. "Your locations" would appear in brand reports. "Across locations" would leak into brand-level analysis.

I documented this as a system rule so any writer, PM, or designer could apply it without asking:

content-system / brand-vs-location.md

Location view

use "your locations"
use "across locations"
use "for all locations"
avoid any brand-level aggregation language

Brand view

use "your brand"
use "for your brand across AI platforms"
avoid "your locations"
avoid "across locations", brand is already the aggregation layer

This isn't a style preference. It's a signal of whether the product understands its own data model. Getting it wrong tells the user the product doesn't know which view they're in.

Terminology standardisation

Search AI surfaces data from multiple AI platforms. Without locked terminology, labels would drift across the product. The same platform could be referenced three different ways across filters, tabs, and chart legends. I defined a single standard and applied it consistently.

content-system / platform-terminology.md
Term Use Avoid Why
AI mode AI mode AI Mode Descriptive category, not a branded feature. Sentence case applies.
Google overviews Google overviews Google Overview Plural, the feature surfaces multiple overviews. "Overview" implies a single fixed artifact.
ChatGPT, Gemini, Perplexity As branded chatgpt, gemini Proper nouns. Always match official capitalisation.

Why this mattered: Terminology drift across a product erodes trust the same way visual inconsistency does. When labels don't match across sections, users question whether they're looking at the same data. Locking these terms meant the product spoke with one voice regardless of which screen you were on.

Building the case for a product category that didn't have established language

The landing page had hard constraints: specific SEO keywords had to land in specific positions, the H1 needed a verb first, and everything had to feel enterprise-credible without sounding like a dashboard feature list.

We ran two variants with different buyer hypotheses:

Variant A: Discovery angle

Dominate local search with an AI visibility checker on ChatGPT

Buyer: Growth / Marketing
Emotion: Discovery

Variant B: Authority angle

Lead local search with an AI visibility checker on ChatGPT

Buyer: Enterprise / Ops
Emotion: Control

The A/B structure wasn't just about tone. It was about learning which buyer converts. Different emotional centers, same structural constraints, same keyword placement.

The product launched. The numbers followed.

Search AI went from inception to beta in 3 months and hit $1M quarterly revenue at launch, called out by the CPO as the fastest-selling product at Birdeye in recent history. The content system built during this period was cited as a direct reason I was recognised as Employee of the Month.

$1M
Quarterly revenue at launch. CPO called it the fastest-selling product in recent history.
3 mo
Inception to beta. Full content coverage from day one: dashboard, LP, email, system.
EoTM
Employee of the Month at Birdeye. Search AI content work cited as a direct reason.
Her shifting to Search AI was instrumental to our successful launch which has delivered tangible outcomes to businesses and Birdeye. With a 3 month launch window, the product quickly scaled to $1M/quarter revenue, making it the fastest selling product upon launch we have seen in the recent past.
Adam Dorfman, CPO, Birdeye (via Slack, March 2025)

What this project taught me about content design at scale

The work that mattered most on Search AI wasn't the headline copy. It was the decisions that prevented drift. The language rules that meant any screen in the product sounded like it was designed, not just populated.

It also reinforced something I believe strongly: content designers need to understand the data model. The Brand vs Location rule only made sense because I understood what the two views were actually measuring. The banner rewrite only worked because I understood the scoring formula well enough to translate it without distorting it.

The best UX copy is invisible. You only notice it when it's wrong.

Figure: reference screenshot from the full case study (embedded image omitted in this web export).