Case Study: What Happens to AI Share of Voice When You Update Content
AI search visibility isn’t trackable the way marketers are used to. For years, SEO performance has been measured through rankings, clicks, and traffic. AI-driven search prioritizes generated answers over traditional link lists, which makes AI Share of Voice a more useful metric for visibility.
AI Share of Voice (AI SoV) reflects how often your brand is cited inside AI-generated answers compared to other sources. When the answer satisfies the query, citations determine who gets seen.
Recently, I conducted an experiment to investigate how the AI Share of Voice (AI SoV) changes when content is updated. To test how a content update actually affects AI Share of Voice, I tracked citation behavior over a 24-hour window.
Here’s what I found.
Key finding: AI SoV shifted within hours of a content update, and continued fluctuating over 24 hours.
What Is AI Share of Voice (AI SoV)?
AI Share of Voice (AI SoV) measures how often a brand or piece of content is cited inside AI-generated answers compared to other sources surfaced for the same prompt.
The calculation for AI SoV is straightforward:
AI Share of Voice = (Brand citations ÷ Total citations in the answer) × 100
For example, if an AI-generated answer cites five sources and a brand appears in two of them, the AI Share of Voice for that prompt is 40%.
Unlike traditional SEO metrics, AI SoV does not:
- Measure rankings
- Assume clicks
- Represent a permanent position
Because AI answers are probabilistic and can change between sessions, AI Share of Voice should be evaluated over time rather than at a single moment.
Why I Ran This Test
AI search is still early. Because it’s so new, we don’t yet have solid answers to some basic questions:
- How do AI systems decide what to cite?
- Why do certain sources show up repeatedly?
- How do small content updates influence how often a brand appears inside generated answers?
What is becoming clear is that AI answers are replacing clicks. Instead of scanning a list of blue links, users are increasingly relying on AI-generated summaries to inform their decisions. In AI search engines, brands are not ranked in the traditional sense—they’re cited.
Based on my research and my work on the Found in AI podcast, a consistent pattern has emerged. AI models tend to favor content that is fresh, structured, and authoritative.
That theme has come up repeatedly in conversations with practitioners working closest to these systems. In one Found in AI episode, Josh Spilker of AirOps mentioned that content often needs to be updated every three to six months to remain competitive in AI search.
That raised a question I hadn’t seen clearly answered anywhere else:
- What actually happens when content is updated?
- Does AI Share of Voice change immediately?
- Do AI systems behave more like traditional search engines, where updates take weeks—sometimes months—to translate into visible movement?
Right now, there are plenty of posts explaining what AI Share of Voice is. But there are very few showing how it actually moves.
This case study was designed to observe movement in real-time and explain why it matters for content marketers and brands trying to appear within AI-generated answers that influence research, shortlists, and purchasing decisions.
What I Was Testing
This AI Share of Voice experiment tested one specific question: Does a fresh content update affect the percentage of an AI-generated answer that a single brand dominates?
To isolate that variable, I kept the scope intentionally narrow. I tested a single prompt across Perplexity: “I need to do a baseline AI visibility audit. Help.”
That prompt consistently returns multi-source answers. It also reflects the type of query content marketers might use when researching AI search visibility. I ran this prompt repeatedly across Perplexity and tracked how often my brand appeared inside the generated answer.
How I Conducted The Test
For the test, I updated one piece of content only: Baseline Audit for AI Search Visibility (Template + KPI Definitions)
No other pages were modified during the experiment.
The primary platform tested was Perplexity. Google AI Overviews were observed incidentally, but they were not part of the structured measurement process. I did not test ChatGPT, Gemini, or other AI models as part of this experiment.
I also did not use any third-party tools claiming to measure AI Share of Voice. Instead, I ran all measurements manually at set intervals. To document the results, I prompted Perplexity, logged the sources that appeared in each answer, and recorded how often each brand was cited.
Because Perplexity typically surfaces multiple sources in a single response, I tracked primary and secondary citations. This allowed me to calculate AI Share of Voice as a percentage of total citations for each snapshot.
To be clear, this experiment is not a:
- Ranking guarantee
- Tool benchmark
- Comprehensive or universal model of AI behavior
This is a controlled, single-page test designed to observe directional behavior on a single platform. The goal was to understand how AI Share of Voice responds immediately after a fresh content update, rather than predicting long-term outcomes across all queries or platforms or producing universal benchmarks.
How AI Share of Voice Was Measured
Before presenting the results, it’s essential to define how the AI Share of Voice was calculated for this experiment and explain why multiple measurements were necessary.
Because Perplexity surfaces multiple citations within a single answer, visibility can’t be measured with a simple yes-or-no mention. Instead, each source needs to be evaluated in context.
For this experiment, citations were counted only when a brand was explicitly referenced in the body of the AI-generated answer. Source cards or sidebar links that were not mentioned in the answer text were excluded.
Citations were categorized as follows:
- Primary citation: A brand or source referenced directly within the body of the AI-generated answer.
- Secondary citation: A supporting source listed alongside or beneath the answer, but not directly referenced in the main response.
AI Share of Voice (AI SoV) was calculated as the percentage of total citations attributed to a single brand within a given answer.
In simple terms, the math looks like this: AI Share of Voice = (Brand citations ÷ Total citations) × 100
This calculation was applied to each snapshot independently.
Why Multiple Measurements Matter
The same prompt can return different combinations of sources depending on timing, context, and how the model weighs available information at that moment.
That variability is why single snapshots are misleading. A one-time check might show a brand dominating an answer—or missing entirely—without reflecting how the system behaves over time.
To account for this, the prompt was run at multiple points over the course of approximately 24 hours. Rather than treating any single response as definitive, these repeated checks were designed to observe movement, range, and central tendency.
Testing Conditions
To mirror real-world user behavior as closely as possible, all prompts were run logged out of Perplexity. This reduced the amount of personalization in any given response.
Measurements were taken:
- On desktop browsers without login
- In private browsing sessions on mobile devices
No personalized or logged-in context was applied. This allowed the experiment to observe how answers changed across devices while avoiding user-specific bias in an attempt to reflect how most users encounter AI answers in practice.
Timeline of Measurements
This experiment was conducted on December 17–18, 2025, following a targeted update to an existing piece of content.
At approximately 8:00 a.m. on December 17, I updated my Baseline Audit for AI Search Visibility blog post. The update included:
- Expanded definitions related to AI search visibility and AI Share of Voice
- Additional guidance on which metrics to track and why they matter
- An updated call to action
- Refreshed author schema
- An updated publish date (from October to December 17, 2025)
Immediately after publishing the update, I submitted the URL for reindexing via Google Search Console.
The goal was to observe how quickly—and in what way—AI-generated answers responded to a fresh content update once it was made available to search and AI systems.
Baseline Snapshot
The first recorded measurement was taken at approximately 8:20 a.m. on December 17.
At this point, the updated content had not yet had time to propagate. Although the page had been published and submitted for reindexing, AI systems had not incorporated the changes meaningfully into the generated answers.
This snapshot serves as the baseline before the update influenced citation behavior.
Post-Update Measurements
After the baseline snapshot, the same prompt was checked repeatedly over the next 24+ hours to observe how AI Share of Voice changed as the updated content was processed.
Measurements were taken at approximately:
- 10:20 a.m. on December 17
- 1:20 p.m. on December 17
- 6:00 p.m. on December 17
- 8:20 p.m. on December 17
- 11:30 a.m. on December 18
Each snapshot recorded:
- Which brands appeared in the answer
- Whether they were cited as primary or secondary sources
- Total citation counts used to calculate AI Share of Voice
To reflect real-world usage, measurements were taken across:
- Logged-out desktop sessions
- Private browsing sessions on mobile devices
Some checks were run logged out, some in private browsing, and some on mobile, which mirrors real user behavior more closely than a single controlled session.
This approach was intentional. AI-generated answers are influenced by context, timing, and system-level weighting. Capturing multiple snapshots across devices and sessions provides a more accurate view of how AI Share of Voice behaves over time.
How AI Share of Voice Changed Over Time
Over the course of this experiment, AI Share of Voice fluctuated significantly.
The baseline measurement, taken at 8:20 a.m. on December 17, showed an AI Share of Voice of 25%. This snapshot represents pre-propagation behavior, before the updated content had been incorporated into AI-generated answers.
Two hours later, at 10:20 a.m., AI Share of Voice increased sharply to 41.67%. At that point, the updated content appeared more prominently within the generated answer.
At 1:20 p.m., AI Share of Voice dropped back to 28.57%, suggesting that early gains were not linear or permanent. This decline was followed by another increase later in the day. At 6:00 p.m., AI Share of Voice rose to 61.11%, the highest observed value during the 24-hour window.
Later that evening, at 8:20 p.m., AI Share of Voice declined again to 40%.
The following day, at 11:30 a.m. on December 18, AI Share of Voice increased once more, reaching 63.16%.
Why a Single Number Doesn’t Tell the Story
Because AI-generated answers change from session to session, it isn’t reasonable to represent AI Share of Voice using a single snapshot or point-in-time metric.
To account for this volatility, I calculated the range, median, and mean across all measurements.
- Range: 25% – 63.16%
- Median: 40.84%
- Mean: 43.25%
AI Share of Voice behaves as a distribution, not a fixed rank. Meaningful evaluation requires looking at movement over time, not simply isolated moments. The median (40.84%) is the most representative single value in this dataset, because it minimizes the influence of short-term spikes and drops.
What This Means
AI Share of Voice can shift meaningfully within hours of a content update. The earliest post-update measurement already reflected a substantial increase in visibility. However, it also reveals something equally important: that movement is not linear.
This experiment also demonstrates that volatility is a normal and, more importantly, measurable phenomenon.
If I had relied on a single snapshot, the conclusions would have been misleading.
The early increase suggested rapid improvement, but later measurements showed pullbacks, followed by additional gains. By 6:00 p.m., more than half of the answer was informed by my content. A few hours later, that influence dropped to 40%. Both moments were real, but neither told the full story on its own.
AI-generated answers are probabilistic, so measuring AI Share of Voice once is not an accurate representation of the overall trend. They shift as systems reweight sources, test combinations, and adapt to context.
Looking at the mean and median provides a far more honest picture of typical visibility than focusing on best-case outcomes. These metrics reflect what users are likely to see across repeated interactions, not just at a single moment when conditions happen to be favorable.
Measuring behavior over time also creates a practical feedback loop. When AI Share of Voice consistently declines across multiple checks, it’s a signal that content may need to be refreshed, expanded, or better aligned with how AI systems interpret authority and relevance.
This is the difference between chasing screenshots and building durable visibility inside AI-generated answers.
If you measure AI visibility once, you’re measuring vibes. But if you measure it over time, you’re measuring behavior.
Why Most AI SoV ‘Case Studies’ Miss This
Most existing research and commentary on AI SoV falls into one of three categories:
- Tool-led success stories: Show outcomes, not how answers shifted over time.
- Vendor dashboards: Present a snapshot without exposing volatility.
- High-level explainers: Define AI SoV, but don’t show how it behaves.
Those resources are useful, but they leave a critical gap. What’s missing from most public examples is behavior.
Very few examples show real-time movement tied directly to a content change. Fewer still document how AI-generated answers shift within hours, rebalance across sources, or fluctuate as systems ingest updated information.
This experiment was designed to observe that behavior directly.
By tracking AI Share of Voice before and after a content update—and measuring it across multiple time points—this case study demonstrates how visibility actually shifts within AI-generated answers. It demonstrates that AI SoV is not static, linear, or accurately represented by a single number.
Importantly, this experiment is repeatable. It isn’t gated behind proprietary tooling or limited to a single brand. Any team can run a similar test across their own prompts and content to understand:
- How quickly AI systems respond to updates
- How often their brand appears inside answers
- When content may need to be refreshed again
That’s the difference between defining AI Share of Voice and understanding how it behaves.
Limitations
This was not a perfect experiment.
Measurements were taken across different devices and sessions, including desktop and mobile, rather than a single tightly controlled environment. The test tracked AI Share of Voice over time for one prompt, one updated page, and one brand, using manual logging rather than automated tooling.
Those limitations matter, and they’re worth naming. At the same time, they’re also what make the signal clean.
By limiting the scope to a single page and a single prompt, the experiment isolates one variable: the impact of a fresh content update.
There were no:
- Competing changes across the site
- Simultaneous updates to other assets
- Tools to smooth or normalize the results
Running the test across devices and logged-out sessions reflects how users actually interact with AI-generated answers in the real world.
AI systems are not experienced through a single, fixed interface. Instead, they’re accessed across contexts, screens, and moments in time. Capturing that variance helps surface behavior, not just best-case conditions.
And within a narrow, controlled scope, the result is clear: fresh, well-structured updates influence how often a brand appears inside AI-generated answers.
The volatility observed doesn’t weaken that conclusion— it reinforces it.
Practical Takeaways for Brands
This experiment has clear implications for how brands of any size should approach AI Share of Voice.
First, it confirms that meaningful content updates can increase AI Share of Voice, especially for brands that already appear inside AI-generated answers. Freshness and structure matter. When these signals are present, AI systems can quickly surface updated content.
But it also makes one thing very clear: measurement cannot be a single moment in time.
Because AI-generated answers fluctuate, AI Share of Voice needs to be tracked across days, weeks, or months. Focusing only on peak visibility creates a distorted picture of how often a brand actually appears in answers users see.
Across the board, brands should:
- Run baseline AI Share of Voice checks before making content updates
- Measure visibility across multiple time points, not single moments
- Prioritize mean and median visibility, not just best-case outcomes
This approach provides a more honest view of how AI systems are surfacing content over time, and when updates are truly having an impact.
A Quick Update Since Running This Test
This analysis focuses on AI Share of Voice movement within the first 24 hours following a content update.
At the time of publishing, Perplexity has begun incorporating an additional article into its answers for this particular prompt. This change occurred after the measurement window documented in this case study and was not included in the original dataset.
Rather than weakening the findings, this reinforces the central takeaway of the experiment. AI-generated answers continue to evolve as new content is discovered, weighted, and re-ranked over time.
Visibility inside these systems doesn’t “lock in” after a single update.
What Comes Next
This case study is not intended to be the final word on AI Share of Voice. It’s simply the first time I’ve slowed down long enough to watch it move in real time.
The next question this experiment raises is straightforward:
What happens to AI Share of Voice over the course of a week after a meaningful content update?
- Does visibility stabilize?
- Does it level out or decline?
- Does it remain consistent as additional content enters the system?
The next phase of this research will explore those questions across multiple prompts and longer observation windows, with findings shared publicly as they become available.
Further reading: The Displacement of Legacy Authority: A 125-Hour AI Share of Voice Case Study
If we haven’t met yet….
Hi, I’m Cassie, a fractional content strategist for early-stage startups who focuses on AI search optimization.
Does your brand need an AI Search Visibility Audit?
My research shows that legacy authority is no longer a shield—Series A/B startups and enterprise brands are being ‘nudged’ out of AI answers by more algorithmically legible competitors. I offer specialized AI search visibility audits to help you identify gaps in your entity coverage and reclaim your Share of Voice.
Request your 7-Day Audit here: AI Search Visibility Audit
What is AI Share of Voice (AI SoV)?
AI Share of Voice (AI SoV) measures how often a brand or piece of content is cited inside AI-generated answers compared to other sources shown for the same prompt.
How do you calculate AI Share of Voice?
AI Share of Voice = (Brand citations ÷ Total citations in the answer) × 100. If an answer includes 12 total citations and your brand accounts for 5, your AI SoV is 41.67%.
What counts as a “citation” in this case study?
Only sources that appeared as citations tied to the answer itself were counted. Sources that appear in side panels or UI modules but are not used as citations in the answer were excluded.
What’s the difference between primary and secondary citations?
Primary citations are sources referenced directly in the body of the AI-generated answer. Secondary citations are supporting sources listed with the answer but not directly referenced in the main response.
Why does AI Share of Voice change between sessions?
AI answers are probabilistic. The same prompt can produce different source combinations depending on timing, system-level weighting, and how the model assembles the response in that moment.
How quickly can AI SoV change after a content update?
This case study observed meaningful movement within hours of a single-page content update. However, results can vary by prompt, topic, and platform.
Is AI Share of Voice a replacement for SEO rankings and traffic?
No. AI SoV is a complementary visibility metric for AI-generated answers. Rankings and traffic still matter, but AI SoV helps you understand how often you’re surfaced when users get answers without clicking.
How should brands measure AI SoV more reliably?
Don’t rely on a single snapshot. Run repeated checks over time (same prompt, same rules) and track mean, median, and range to understand typical visibility rather than peak moments.
Can I replicate this experiment for my own brand?
Yes. Pick a single prompt, define what you’ll count as a citation, run logged-out checks at set intervals, and track how often each brand is cited. Keep the scope narrow so you can isolate the effect of updates.
What are the limitations of this case study?
This was a single-prompt, single-page test on one platform with manual logging across sessions/devices. It’s designed to show directional behavior, not provide universal benchmarks or guarantees.