Citation Decay
Citation Decay is a measurable decline in an entity's observed citation rate between comparable measurement windows. It records a downward change in AI citation presence — separate from organic search rankings, claim support, or business outcomes.
Defined at the home of Machine Relations: Citation Decay in the Machine Relations glossary
Citation decay is something you measure, not something you predict. A brand shows it when the same questions, asked of the same engines under the same conditions, cite the brand less often in a later window than in an earlier one. The definition and the full measurement protocol live in the Machine Relations glossary. This entry is about reading it inside an earned-media program.
Turnover is not decay
Engines change their sources from one answer to the next. In the Machine Relations Index's 31 October 2026 release, pooled across five engines, a domain cited for a question was cited again by the same engine for the same question 51.6% of the time within 7 days and 21.7% of the time 92 or more days later. For the exact page, the figures were 46.0% and 14.9%. Perplexity comparisons that span its 2026 source-count changes are left out (AI citation repeat rates).
That is source turnover across every cited domain. It shows how much a single brand's citations can move with nothing wrong at all. A drop has to clear that movement, on repeated runs, before it is decay.
How to tell the two apart
- Fix the panel: the questions, the engines and interfaces, location and account state, and the rule for what counts as a citation.
- Collect repeated answers in two comparable windows. One answer is a sample.
- Compare the brand's citation rate in the later window with the earlier one, engine by engine.
- Read the change against that engine's turnover on the same questions.
- Call it decay only when the lower rate holds. A fall in one engine says nothing about another.
If the protocol changed between the windows (new questions, a different interface, an engine changing how many sources it retrieves), the windows are not comparable. Start a new baseline instead of reporting decay.
What a decline does not tell you
- Why it happened. A lower rate records what changed. The cause is a separate question.
- That a page died. A cited address that now returns 404 is a lost page, not a lower citation rate. Of 1,043 measurable pages engines last cited between July and October 2026, 3.3% returned 404 or 410 when fetched on November 10 (dead cited pages study).
- An SEO ranking loss. A page can hold its search position and be cited less, or the reverse. Record the two separately.
What to check when the decline is real
These are things to test, not causes the decline establishes:
- The cited pages. Still live, still saying what was cited, still the clearest answer to the question.
- The replacements. Which sources the engine cites instead, and when they were published.
- The question and the engine. Whether the wording buyers use changed, or the engine changed how it retrieves.
- The coverage behind the brand. Whether independent articles that named it were removed, paywalled or blocked to crawlers.
For an earned-media program, the useful habit is to log placement dates beside the panel. A decline can then be read against what changed, instead of being explained after the fact.
Frequently asked questions
How quickly do AI citations decay?
There is no general timeline for a brand. What the Machine Relations Index measures is turnover: about half of the domains cited for a question are cited again by the same engine within a week, and about a fifth after three months or more. A brand's own decline has to be measured on its own repeated, comparable runs.
Does pausing PR cause citation decay?
A decline alone cannot establish a cause. To test it, keep a fixed panel, log when placements ran and when they stopped, and compare the windows.
Does updating a page stop citation decay?
Changing a date without changing the content gives an engine nothing new to cite. If you update a page, record the change and compare later runs on an unchanged panel.
How is citation decay different from citation velocity?
Citation velocity records citations that appear where the comparable earlier observation had none. Citation decay records a lower citation rate between comparable windows. Report them as separate measurements.
See how your brand performs in AI search
Free AI visibility audit: your AI visibility score, the publications already covering you, the gaps, and the story angles to close them.
Run Free Audit