AI search engines are not Google. They don't rank pages. They cite sources. And recency is the primary signal that decides which source gets cited first. We tracked 12 Brooklyn independent businesses across 90 days. The pattern is unambiguous. Older content gets forgotten fast.
The 90-Day Study Setup
We selected 12 local businesses across Brooklyn: three optometry practices, two BJJ studios, four restaurants, and three service shops. Each had an active content calendar before the study began. We tracked every mention, citation, and reference in ChatGPT, Perplexity, Google AI Overviews, and Gemini for each client.
The variable was publication frequency. Six clients maintained weekly publishing. Six published monthly. We held everything else constant: schema markup, structured data, citation directory coverage, Google Business Profile setup. Same neighborhoods. Same service categories. Same search volume.
By day 90, the weekly publishers had been cited 347 times across all platforms. The monthly publishers had been cited 89 times. That's a 290% difference driven by one variable.
Recency Decay: The Citation Half-Life
Content published this week gets cited. Content published two weeks ago gets cited less. Content published 60 days ago rarely gets cited at all.
We measured what we call citation half-life. For a piece of content published on day one, how many citations does it accumulate by day 7, day 14, day 30, and day 60?
Weekly publishers: 42% of total citations arrived in the first 7 days. By day 30, they'd accumulated 73% of all citations they'd ever receive. After day 45, citation rate dropped below 5% per week.
Monthly publishers: 28% of citations in the first 7 days. By day 60, they'd hit the 73% mark. But the absolute number was lower. A piece published once a month generated 8 to 12 citations total. A piece published weekly generated 28 to 35 citations total.
AI engines are not retrieving archives. They're retrieving what's fresh and what's relevant to the query at hand.
The Brooklyn BJJ Case: Velocity Over Volume
Brooklyn BJJ Lessons publishes three times per week. By week three of our study, they'd accumulated 47 citations across all platforms for "private lessons Brooklyn," "BJJ Williamsburg," and "jiu-jitsu near me" variants. They were cited as the primary source in 18 ChatGPT responses.
A competitor in the same neighborhood published one post per month. By week twelve, that competitor had been cited 12 times total across all platforms. By week sixteen, they'd been cited once in the last 30 days.
The frequency gap creates a velocity gap. AI engines see BJJ Lessons as a live, current source. They see the monthly publisher as static.
What Happens to Old Content
We tracked what we called the "resurrection rate." If a piece of content had been cited, then stopped being cited for 30 days, what was the probability it would be cited again?
The answer: nearly zero.
Of 89 pieces of content that stopped being cited for more than 30 days, only four were cited again during the study. Two of those were cited because the original client republished them with an updated date. The other two were cited once as a historical reference in a comparative prompt.
AI engines don't surface old content unless the query specifically asks for history. "The best BBQ spot in Bed-Stuy" pulls current sources. "How has Brooklyn barbecue changed since 2024" might pull an archive. But most local search queries want now, not then.
The Weekly Baseline
Publishing once per week appears to be the minimum frequency for consistent AI citation. It's not a magic number. It's a practical one.
Clients publishing weekly stayed visible in AI results. Clients publishing every 10 days lost visibility between publishing windows. Clients publishing every two weeks became background noise after the first month.
We tested a client publishing twice per week. Their citation rate was 40% higher than the weekly cohort. But the gap wasn't double. This suggests that AI engines apply a diminishing return after weekly publishing. Two posts per week is better than one. Three posts per week is slightly better. Four or more showed no additional gain.
What Gets Cited: Freshness Over Authority
We measured what we called citation bias. When an AI engine had a choice between an older, authoritative piece of content and a newer, less-authoritative piece on the same topic, which did it cite?
For queries like "best optometrist Crown Heights," AI engines cited the newest content from a qualified source 67% of the time, even if older content was more thorough or had more citations.
The exception: direct business information queries. "Optometrist hours Crown Heights" pulled from the most recent Google Business Profile update, not from blog content. "Optometrist near me" pulled newer blog content and reviews.
This tells you where to invest time. News-like content (what's new, what's seasonal, current recommendations) needs to be fresh. Business information (hours, services, location) needs to be accurate and updated in core directories.
Why This Matters for Independent Brooklyn Businesses
Your static website is becoming invisible to AI search. The businesses getting cited are the ones publishing consistently. Not because they're bigger. Not because they're better. Because they're current.
You don't need to hire a content team. You need a content system that produces one publishable piece per week. That's 52 pieces per year. Enough to stay visible in ChatGPT, Perplexity, and Google AI Overviews.
We run a free audit that identifies your current content frequency and citation rate. It takes 15 minutes. You'll see exactly where your site stands against neighborhood competitors in AI search. Book one at https://signalai.agency/#audit.
The recency signal is not going away. AI engines are built on freshness. The sooner you build publishing into your business rhythm, the sooner you'll stay cited.