Choose one service that you currently offer and its page. Then collect customers’ search phrases and one or two questions they might ask an AI system. A technical check, search positions, and AI answers will show different sides of the same task.
Suppose you install ventilation systems in offices. The page should load and explain the terms, search results should lead to the relevant URL, and you can check AI answers for mentions of the company and accurate information. Start with one offering so you have a clear basis for comparison.
Choose a page for a current service
Start with an offering that the company is actually ready to provide. Write down the service name, service area, key terms, and the action you expect visitors to take. For example: request a quote after an inspection of the site. If the service is not part of your current offering, tracking a promising phrase will not give the team a clear task.
Choose the current URL of the page. Open it without signing in, read its content, and check that the contact details or form relate to this service. Record the final URL after any redirects. This is the URL to compare against the audit and the search results you find.
Keep a brief description of the customer nearby: what kind of property they have and what they want to find out. This helps distinguish an installation page from an equipment maintenance page. Both may be technically sound but address different needs. For help choosing URLs to monitor, see the guide to pages for monitoring.
Prepare phrases and questions
A search phrase can be short: “office ventilation installation.” An AI question usually states the task more explicitly: “Which companies install ventilation systems in offices in this city?” Save both exact texts. When reporting the results, do not replace them with the general word “ventilation.”
For search queries, specify the region and frequency group. Up to fifty queries fall into the daily group in Yandex and Google; up to another one hundred and fifty fall into the weekly group, and up to another eight hundred into the monthly group in Yandex. Choose the frequency based on your work needs, not the number of similar phrases.
AI questions are checked monthly across seven services. You can save from one to five of your own unique questions, each up to 250 characters long. A new wording takes effect at the next check; saving the setting does not mean that a new complete set of answers is returned immediately.
Record the check conditions
The daily automated audit is designed for sites of up to one hundred pages. Check whether the selected service page was included in a specific report. If it was not, do not present the project’s overall result as a technical check of that service. If necessary, run a manual audit of the relevant pages and record its scope.
For a position check, use the date of the actual available measurement, the URL found, the search engine, and the conditions. For AI, record the question text, service, date, and answer status. Some data may still be collecting or may be unavailable. A missing result is not a zero mention or a lost position.
You can record results for one service in a separate table:
| Source | What to record | What question it helps answer |
|---|---|---|
| Service page audit | URL, date, specific finding | Is there a confirmed technical obstacle? |
| Search measurement | Phrase, position, page found | How does the site appear for this phrase? |
| AI answer | Question, service, answer, mention | Was the company named in this observation? |
Check the page technically
If the audit finds an issue, open the details and verify the underlying fact. The page may return an incorrect status, contain an indexing restriction, or point its canonical (the URL of the primary version of the page) to a different service. Choose a finding that relates to the current URL and state the expected outcome.
Give the developer the URL, the check result, the steps to reproduce it, and the criteria for verifying the outcome. Do not add a requirement to “rank first in search after the fix”: technical work should have a verifiable technical outcome. Confirm successful publication with a new response from the page and a repeat check covering the relevant scope.
If no significant technical obstacles are found, move on to the content and offering. Check whether the page answers questions about installation, explains the service area and scope of work, and gives visitors a next step. The absence of an error in the report does not prove that the content is useful enough for the customer.
Review the search result
If a position has dropped, first make sure the measurement points are comparable: the same phrase, region, device, and search engine. Then look at the page that was found. If a general section appears instead of the service page, investigate differences in content and internal links, and record both URLs.
Comparing a competitor is useful when you examine a specific pair of pages for a query. Look at the terms and information on each page. A competitor’s longer text or larger number of headings does not establish why it ranks where it does. Choose content improvements based on the customer’s needs, not by copying metrics.
The guide to building a keyword set explains how to develop a set of phrases. For a specific drop, see the guide from query to page check. Both tasks require the actual URL found, not just the service page you wanted to appear.
Read the AI answers
Open the result from a specific service for the saved question. Separate a mention of the company, a link to its website, and the factual accuracy of the description. If the company is named but the service is described incorrectly, that is a different issue from no mention at all. An aggregate number is no substitute for reading the answer.
A zero result means no mention was detected in the answers received for that check. It does not prove that AI systems never mention the company. If an answer is incorrect or incomplete, record its status as unknown and do not automatically count it among negative results.
If you find outdated information, check your own accessible pages and the sources cited in the answer. You can separately confirm that information on the website has been corrected, but you cannot promise that the next AI answer will immediately reflect the new version. The service and the website have different update processes.
Choose one fix
For the example service, the task could be: “Update the installation page: correct the canonical and clearly state which office properties we serve.” Separate the technical fix from the editorial work; assign owners and check the published version.
After publication, confirm the URL’s status. Then monitor the next available position for the same phrase and the next monthly AI answer for the same question. Note in the log that other factors may have affected the results between measurements. An improved observation does not prove that a single change caused it.
You can organize this workflow in the site monitoring subscription: three sources provide context for one work task. The subscription is useful for regular checks and connecting results, while the team chooses what to do based on confirmed facts. This makes it possible to fix the site and explain the work without conflating an audit, a search position, and an AI answer into a single notional score.