AI MONITORING
Answer engine responses change even when we do nothing. So a single measurement is not a result but one sample. We repeat the baseline the audit created with the same questions under the same conditions, and record what shifted in between.
In one paragraph
Citation monitoring is the operational work of repeating the baseline an audit created under the same conditions. Holding question wording, target engines, session conditions and the aggregation denominator fixed, we ask the same questions again weekly and record mention rate, citation rate, recommendation appearance and competitor share separately. Sentences answered contrary to fact get a correction evidence document, tracked until they are cited again. Because a single measurement cannot speak to change, fixing the conditions comes before measuring.
What this service sells is not numbers but comparability. Put measurements taken under slightly different conditions side by side and it looks like a trend, while in fact they measured different things. We only start re-measuring after the conditions are frozen in writing, and when a condition has to change we state in the report that it changed.
Last verified
WHEN YOU NEED THIS
These are the situations we hear repeatedly in consultations. If any of them apply, start by measuring.
The reason splits three ways — our document was pushed out, another domain took the slot, or the engine refreshed its index and changed the composition of the answer itself. Telling those three apart requires several rounds under the same conditions.
With no baseline before starting, or with a baseline re-measured using different questions, the number has nothing to compare against. Re-measurement has to be a repetition of the baseline, not a new measurement.
The moment a correction document goes up and the moment the answer actually changes are different. You have to record the incorrect sentence as it was and ask the same question again to confirm.
When the denominator moves, the citation rate becomes a result of arithmetic rather than of work. Fixing the question count and the engine count and stating them in every report is what makes it readable as a trend.
WHAT WE DO
Written as units of work rather than abstract proposals. The scope of an engagement is set from this list.
The conditions used for re-measurement are fixed in writing. The question set settled in the audit carries over exactly, and after that not one character of the wording changes.
Mention, citation, recommendation and share are counted apart rather than merged. Each metric's denominator is stated in every report so the number never travels away from the conditions it came from.
All 7 answer engines are recorded separately. Merging them into one average hides a change that started in a single engine, which means missing the earliest signal.
A wrong answer is not closed at discovery; it is tracked as one item through to confirming re-citation. Where the audit stops at listing wrong answers, monitoring checks each round whether the answer actually changed for the same question after the correction document went up.
Work history — publications, structural changes, index submissions — sits beside the measurement record. It is so that when a change is observed, the first thing established is whether it came after our work or before it.
Weekly observations get bundled monthly into trends and next actions. Not only the items that improved — the ones that got worse and the changes that cannot be explained go in as they are.
PROCESS
What you receive at each stage is stated alongside it. Durations are the working time Navirang controls; they are not a promise about when results appear.
The audit report's question set and observation record are confirmed as the baseline as they stand. If the audit was done elsewhere, we first check whether the conditions can be reproduced.
Baseline confirmation
Target engines, cadence, session conditions, metric definitions and the denominator are fixed in writing. Every ratio in later reports is only read under the conditions in that document.
Measurement conditions document
The fixed questions are asked again under the same conditions, recording mentions, citations, recommendations and competitors. Measurement is read-only; we never intervene in the result.
Weekly observation record
When a wrong answer is observed we build the evidence document, restate the fact in structured data and submit for indexing — then keep confirming re-citation with the same question.
Misinformation correction history
A month of observations organised into trends and cross-referenced against work history. Changes that cannot be explained are written as not explained.
Monthly citation report
DELIVERABLES
We do not do work that ends in conversation. The documents below remain, and become the baseline for the next measurement.
Question wording, target engines, session conditions, metric definitions and the denominator, fixed. Delivered before any number, with a change history if it moves.
The same cells (question × engine) refilled each round. Each cell carries mention, citation, recommendation and competitors, with the round number and timestamp attached, so what flipped since the previous round is visible in the table.
Mention rate, citation rate, recommendation appearance and competitor share by round, with the denominator (questions × engines) stated beside each value.
A record joining the incorrect sentence, the engine, the correction evidence document, the index submission date and the re-citation confirmation date in one line.
Trends, change per engine, the cross-reference against work history, and next month's items to check. Items that got worse carry the same weight.
The same rule is applied to our own site — 40 questions × 7 answer engines, a denominator of 280 cells. That denominator does not change once the first round is recorded, because changing it makes comparison with earlier rounds impossible; if a new engine is worth counting we start a separate set rather than editing the existing one. Baseline measurement is still running so there are no results yet, and when there are they will be published with the full question set, the measurement protocol and the raw CSV.
HOW IT CONNECTS
Our work moves as one piece. SEO builds the foundation for being found by search engines, AEO raises the odds of that information being cited in an answer, structured data helps machines understand the facts, and content supplies the evidence there is to cite.
| Area | Relationship to this work |
|---|---|
| AEO audit | The single measurement that creates the baseline monitoring repeats |
| Citable content design | Building documents to answer the questions monitoring keeps finding empty |
| Structured data | Making the corrected fact read as the same value on the machine side |
| Brand marketing | Where the cause of a wrong answer is brand notation or entity inconsistency |
| Indexing | Submission and index status checks that get the correction document read again |
| AEO by industry | Re-measuring an industry question set under the same conditions |
FAQ
A The audit is a single measurement; monitoring is repetition under the same conditions. The audit finds what the current citation status is and why, creating a baseline; monitoring repeats that baseline with the same questions, engines and session conditions to track change. Monitoring without an audit has nothing to compare against; an audit without monitoring cannot confirm whether anything improved.
A Four, counted apart. Mention rate is answers where the brand name appeared in the text divided by total answers measured; citation rate is answers where our domain was used as a source over the same denominator. Recommendation appearance is the proportion of vendor-search questions where we were offered as a candidate, and competitor share of voice is each brand's share of appearances across the same question set. We do not merge the four into one score, because that makes it impossible to trace back what improved and what got worse.
A Weekly, consolidated into a monthly report. What matters more than the frequency is that the conditions repeat. Answer engines give different answers to the same question depending on the moment and the session, so a single round is treated as one sample and judgement comes from the trend.
A We cannot assert it, which is why we hand over the material for judging. Engine-side refreshes happen in the same period, so an observed change is not attributed straight to the work. We put the history of publications, structural changes and index submissions beside each measurement round so you can see what a change followed; items that moved in stretches where we touched nothing are marked separately; and changes that cannot be explained are written as not explained.
A Not immediately. Some engines have a reporting channel, but it does not function as a place that corrects an individual fact on request, so we proceed by changing the evidence document and getting it read again. We build the correction page, restate the fact in structured data, submit for indexing, and track it until it is cited again with the same question. Reflection speed differs per engine — an engine leaning on live search can reflect a newly indexed document within days, while an answer coming from learned knowledge takes far longer.
A We cannot. What decides citation is neither the agency nor the site owner but the answer engine, and because the same question gives different answers depending on the moment and the session, being cited in one round does not mean the next will match. What monitoring does is not prevent a drop-off but confirm, without delay, which engine and which question it started from. What we promise is not a result but a scope of work and a measurement method — what we measure, under which conditions, and how we report it, given in writing before the contract.
A No. Automatically asking the same question repeatedly, or auto-clicking your own results, is abuse, and our measurement runs read-only. If the act of measuring affects what is being measured, the record loses its value as evidence.
With no baseline, start with the audit; with one already, we check the conditions and take it over as it stands. The conditions document comes first, then the numbers.
We reply within one business day.