Last updated July 2026.
Most GEO dashboards tell you how often your brand appears in AI answers. Fewer tell you which specific URL was cited, how it was framed, and whether that citation survived the next engine update. Without that detail, you are flying blind on content decisions.
The tracker below fixes that. You can run it in a Google Sheet, Notion database, or Airtable. The schema is stable enough that platforms like Profound, Peec AI, Otterly.AI, and Temso can feed data into it automatically once you have the columns set.
The core column schema
Copy these 11 columns as the header row of your tracker.
| Column | What to log |
|---|---|
| URL cited | The exact page URL that appeared as a source, not just the root domain |
| Engine | ChatGPT, Perplexity, Gemini, Google AI Overviews, Microsoft Copilot, or other |
| Prompt cluster | The query or topic group that triggered the citation (e.g., “best GEO tools”, “how to track AI citations”) |
| First-cited date | The date you first confirmed the citation, not the page publish date |
| Framing type | Primary recommendation / list item / passing mention (see framing guide below) |
| Re-check 30d | Date of 30-day re-check |
| Still cited 30d | Y or N |
| Re-check 60d | Date of 60-day re-check |
| Still cited 60d | Y or N |
| Re-check 90d | Date of 90-day re-check |
| Still cited 90d | Y or N |
| Notes | Page changes, content updates, or any structural shifts between checks |
Framing types: why they matter
The same page can appear in an AI answer three different ways. Track which framing you get, because framing predicts stability and conversions differently.
Primary recommendation is when the engine names your page or domain as the top answer for the query. This is the strongest position and tends to have the longest half-life. Example: “For tracking AI citations, [domain] provides the most complete method.”
List item is when your page appears as one of several sources in a structured answer. This is the most common framing. Citation half-life is shorter here because engines rotate list members more frequently as new content enters their index.
Passing mention is when your domain or page is referenced in context but not as a direct source recommendation. This framing has the shortest half-life and the lowest click-through value.
Log the framing every time you re-check. A citation that downgrades from primary recommendation to passing mention is a leading indicator of decay, even if the page technically still appears.
The citation half-life formula
Half-life = log(0.5) / log(retention rate per interval)
Where the retention rate per interval is the share of citations that survive one re-check period.
Example: You log 20 URLs as cited on Day 0. At the 30-day re-check, 16 are still cited. The retention rate is 16/20 = 0.80.
Half-life = log(0.5) / log(0.80) = (-0.301) / (-0.097) = approximately 93 days
That means you would expect half of your cited URLs to have rotated out of AI answers within about 93 days. Update your content, earn new mentions, and re-check those URLs before the half-life threshold hits.
How fast do citations appear and decay?
Two data points inform the cadence, with important caveats on both.
On appearance speed: a figure of roughly 6.81 days as a median time from content publication to first citation has circulated in practitioner discussions attributed to Profound tracking approximately 900 marketing pages between March and May 2026. That figure has not appeared in a published Profound study at time of writing and should be treated as an informal benchmark rather than a confirmed research finding. What is more robustly established is that AI engines index and cite new content measurably faster than traditional organic ranking cycles, so a 14-day minimum before your first citation check is a reasonable starting assumption.
On content age and citation survival: a Seer Interactive vendor study of approximately 5,000 URLs with extractable publish dates found that roughly 85% of Google AI Overview citations came from content published between 2023 and 2025, suggesting that AI Overviews skew toward content published in the prior two years. That finding is contested: a separate, larger Ahrefs study of 17 million citations found that AI Overviews actually cited older content on average than other AI platforms, making it the least freshness-biased platform tested. The two studies used different methodologies and different platform scopes, and neither should be taken as settled fact. What both studies agree on is that content age interacts with quality and authority signals in ways that are not yet fully predictable.
The practical read: log the first-cited date for every URL, run your 30/60/90-day re-checks, and let your own half-life data replace generic benchmarks as quickly as possible.
Worked example rows
Here is a sample tracker populated with three fictional but realistic rows to show the schema in practice.
| URL cited | Engine | Prompt cluster | First-cited date | Framing type | Still cited 30d | Still cited 60d | Still cited 90d | Notes |
|---|---|---|---|---|---|---|---|---|
| /blog/geo-tools-2026 | ChatGPT | ”best GEO tools” | 2026-05-01 | List item | Y | Y | N | Page not updated. Competitor published a newer round-up in June 2026. |
| /guides/citation-share-formula | Perplexity | ”how to calculate citation share” | 2026-05-10 | Primary recommendation | Y | Y | Y | Direct-answer paragraph at top of page. No change needed. |
| /tools/temso-review | Google AI Overviews | ”GEO platform reviews” | 2026-06-01 | Passing mention | N | N/A | N/A | Citation lost at 30d. Page has no FAQ schema and buries the verdict. Flagged for rewrite. |
Row 1 illustrates a list-item citation that survived two checks but fell at 90 days after a competitor published fresher content. Row 2 shows a primary-recommendation citation that held across all three re-checks because the page leads with a direct-answer paragraph. Row 3 shows an early citation loss at 30 days, with the notes column capturing the structural cause.
The pattern these rows reveal: primary-recommendation framing with a direct-answer opening paragraph appears to hold longer than list-item or passing-mention framing. That matches what the retrieval layer of most AI engines favors, based on how they surface concise, quotable passages.
How to fill the tracker automatically
Manual spot-checks are a viable starting point for prompt sets under 20 queries. For anything larger, you need a tool that surfaces the cited URL, not just the domain.
Profound provides citation maps that show which specific page appeared in an AI answer and for which prompt, across nine or more engines. That data maps directly into the URL and Framing Type columns. Entry price is $99/mo for ChatGPT-only coverage, or $399/mo for full engine coverage.
Peec AI offers granular citation tracking across Google AI Overviews, ChatGPT, Gemini, and Perplexity, with the ability to export cited URLs per prompt. Useful for teams whose primary concern is Google AI Overviews alongside the chat engines.
Otterly.AI tracks citations at the prompt level across six platforms. The GEO Audit Engine checks individual URLs for citation readiness, which complements the re-check columns in the tracker. Entry is $29/mo for 15 prompts, scaling to $189/mo for 100 prompts.
Temso monitors citations across eight engines from a single subscription starting at $89/mo, surfaces citation gaps, and flags which specific pages are and are not being cited for a given prompt cluster. It covers all three steps of the GEO loop (track, diagnose, execute) without per-engine add-on fees, which makes it a practical option for teams managing a large prompt set across multiple engines. It is not a substitute for a traditional SEO suite but works well as the citation layer.
For the re-check workflow specifically: set a recurring calendar alert at 30, 60, and 90 days after each new entry. If you use Profound or Peec AI, run the same prompt cluster on the same engine at each interval and compare the cited URL to what you logged. If you use Otterly.AI or Temso, export the citation list for that prompt cluster and check whether your logged URL still appears.
The sourced stat box: what we know about citation appearance and decay
| Claim | Source | Confidence | How to use it |
|---|---|---|---|
| Roughly 85% of Google AI Overview citations come from content published in the prior two years | Seer Interactive vendor study of approximately 5,000 URLs (2025) | Low: contested by a larger Ahrefs 17M-citation study that found AI Overviews skew toward older content | Use as one data point; do not treat as a universal freshness rule |
| AI-cited content is on average 25.7% fresher than Google-organic-ranked content | Ahrefs (2025, 17 million citations) | Moderate: large dataset, but a correlation, not a causal study | Pages you update have a modest freshness advantage over pages left static |
| Median time-to-first-citation of roughly 6.81 days (Profound, March-May 2026) | Circulated in practitioner discussions; no published Profound study confirmed at time of writing | Unconfirmed: treat as informal benchmark only | Use 14 days as a conservative starting assumption for your first citation check |
| Only 11% of domains are cited by both ChatGPT and Perplexity | Profound analysis of 100,000 prompts (July 2025) | Partially verified: vendor research, no independent replication | Reinforces why per-engine columns matter in your tracker |
What to do with the data
After 90 days of re-checks, you have three actionable outputs.
Half-life by engine. Calculate the retention rate and half-life separately for each engine column. A short half-life on Perplexity but a long one on Google AI Overviews tells you that your content is structurally stable for one engine’s citation patterns but not the other’s. Perplexity typically skews toward high-authority domains and communities; AI Overviews skew toward structured, on-domain content.
Framing distribution. Count how many citations are primary recommendations versus list items versus passing mentions. If the majority are passing mentions, your pages are being pulled into context but not ranked as the top answer. That is a structural problem in the opening paragraph, not a content freshness problem.
Decay cohorts. Group pages by first-cited month and track how each cohort’s retention rate changes over time. A 2026 cohort with higher retention than a 2025 cohort tells you that recent content updates are paying off. The reverse tells you that older pages have accumulated authority signals that newer content has not yet earned.
The tracker operationalizes what citation share dashboards in tools like Profound, Peec AI, Otterly.AI, and Temso measure at the platform level. The platform gives you the number; the tracker gives you the story behind why that number is moving.
Start tracking today
Copy the 11-column schema into a Google Sheet, add your first five cited URLs from a manual spot-check in ChatGPT or Perplexity, and set a 30-day calendar alert. That is enough to generate your first half-life data point.
For the full picture of which GEO tools can automate this workflow, see the GEO tool ranking and the GEO glossary for definitions of citation share, share of voice, and related terms. The methodology page documents how this site scores and evaluates the tools referenced above.