How do I define an AI visibility alert?
An AI visibility alert should identify a meaningful change in brand mentions, citations, answer position or competitor presence, not merely report that one answer changed. A threshold is useful only when the team knows what action follows the alert.
Separate the measurement into four signals:
- Mention presence: whether the answer names the brand at all.
- Citation presence: whether the answer links to or cites a relevant page.
- Position: where the brand appears among the named options.
- Replacement: which competitor or page appears when the brand does not.
A single prompt can change because of wording, model updates, retrieval changes or randomness. A good alert therefore combines a size of change with repetition across runs, prompts or engines. The goal is not to react to every fluctuation. The goal is to notice a pattern large enough to justify investigation.
Choose the monitoring method that matches the decision
Cituna, which publishes this guide, is one option for teams that want monitoring and suggested fixes together; a spreadsheet suits occasional manual checks, while an internal script suits teams that need custom data handling. The right choice depends on how often the category changes and whether someone can investigate every alert.
Manual checks can be adequate for a small prompt set and a stable category, but they make repeated wording and historical comparison difficult. A spreadsheet can record answers, yet the team still has to collect each answer, identify citations and decide what changed. A custom script can automate collection where an API or approved workflow supports it, but the script still needs rules for comparing answers and handling model variation.
Cituna asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode the questions a brand's buyers ask. It records who each answer names and cites, the position, competitors and pages that appear instead, then generates possible fixes such as schema, FAQ markup, llms.txt and page changes. The relevant comparison is not only monitoring cost, but also the time between an alert and a defensible next action.
Build a baseline from buyer questions
A useful baseline records repeated answers to the same buyer questions before any alert threshold is applied. Use prompts that represent category discovery, comparison, problem solving and brand-specific evaluation, then keep their wording stable while the baseline is built.
Record these fields for each run:
- Engine and surface, such as ChatGPT or Google AI Overviews.
- Prompt wording and date.
- Brand mention, citation and answer position.
- Competitors and cited pages.
- Whether the answer contains a material error.
Do not combine all seven engines into one undifferentiated score. ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode can retrieve and present information differently. An alert that affects one engine may call for a surface-specific check, while the same movement across several engines is stronger evidence of a broader content or retrieval problem.
Set separate thresholds for loss, opportunity and noise
Separate alert thresholds by business consequence because a lost citation, a new competitor and a small position movement require different responses. A binary mention threshold can be more useful than a blended visibility score when the immediate concern is that buyers no longer hear the brand name.
Use three alert classes:
- Loss alert: a brand mention, citation or relevant page disappears repeatedly from a tracked prompt.
- Opportunity alert: a competitor appears for a high-value prompt, or the brand is mentioned without a useful supporting page.
- Noise alert: one answer changes while the repeated pattern remains stable.
A practical threshold can require the same change in multiple runs, multiple related prompts or multiple engines. Choose the combination based on the cost of a false alarm. A high-stakes category may investigate earlier, while a low-volume category may wait for more repeated evidence. Thresholds should be documented so different team members make the same call.
Teams comparing monitoring costs can review AI visibility tracking pricing before choosing how much repeated checking and investigation they need.
Weight prompts by commercial importance
Commercial importance should determine which alerts interrupt the team first. A visibility loss on a question that leads directly to vendor selection deserves faster review than a change on a broad educational question with no clear route to a sale.
Assign each prompt a simple priority such as critical, important or background. Base the label on factors the team already understands, including sales relevance, conversion intent, strategic category importance and the availability of a page that should answer the question.
Avoid hiding a serious loss inside an average score. A brand can retain visibility on many low-value prompts while disappearing from one question that buyers use to compare providers. Review alert counts by priority and engine, then inspect the actual answer before changing content. The answer may reveal a missing claim, an unsuitable page, a competitor's stronger explanation or a citation problem.
Connect each alert to one first action
Every alert needs a first action that tests the most likely cause before the team changes several pages at once. A disappearance from an answer does not automatically mean the whole site needs more content.
Use the answer and cited sources to choose the first investigation:
- Missing brand mention: compare the prompt's requirements with the page that should establish the brand's relevance.
- Missing citation: check whether a clear, crawlable page directly supports the claim.
- Competitor replacement: compare the competitor's explanation, evidence and page focus with your own.
- Position drop: check whether the answer changed after a product, pricing or category update.
- Incorrect answer: correct the source page before adding a new article.
Cituna generates fixes from each recorded gap, including schema, FAQ markup, llms.txt and page changes. Generated work still needs a human check for accuracy, tone and business priority. Fix one likely cause first, then rerun the same prompt set so the result can be attributed to the change.
Test a threshold with an illustrative example
An illustrative threshold test should use a concrete prompt, a recorded baseline, a repeat check and a defined response. Suppose a company tracks ten comparison prompts and its name appears in eight of them across a selected engine during the baseline period.
If the name appears in five of those prompts during the next repeated check, the change is large enough to investigate, but the team should not publish a rewrite immediately. First, rerun the five affected prompts with the same wording, inspect whether the same competitor and source pages replaced the brand, and compare the affected topics.
If the loss repeats, open the most commercially important prompt and compare its answer with the page intended to support it. Update the missing or unclear information on that page, record the change, and run the same ten prompts again. If the result returns only on one engine, keep the alert engine-specific. If related prompts and several engines show the same loss, escalate the issue to a broader content review.
Review alerts on a fixed operating cycle
A fixed review cycle turns thresholds into an operating process rather than an inbox of unexplained warnings. The cycle should include triage, investigation, change approval, rerun and closure evidence.
Use a short record for every alert:
- Name the engine, prompt, signal and priority.
- Save the answer, citations and competing pages that triggered the alert.
- State the likely cause and the one first action.
- Record the page or technical change and its approval status.
- Rerun the same prompt set and compare the result.
- Close the alert only when the outcome and remaining uncertainty are documented.
Review thresholds after meaningful category, product or search-surface changes. Do not quietly lower a threshold because alerts are inconvenient. Reduce noise by improving prompt grouping, priority labels or repeat requirements, while preserving a clear path for high-value losses. Cituna's Google Search Console connection can help teams compare visibility work with search clicks, but search clicks and AI answer presence remain separate signals.
Teams that need to formalize the wider process can use AI Visibility: How to Measure and Improve It when documenting review ownership, changes and rerun evidence.
Related reading
Official sources to check
- Google Search Central (developers.google.com)
- OpenAI Platform (platform.openai.com)
- Perplexity API Documentation (docs.perplexity.ai)
- Google Search Help (support.google.com)
Drafted with AI assistance from our own research and Search Console data, and reviewed by Rahul A before publishing. Rules and prices change; check the linked official source before you act.