Aizii Learn · AI Search
← Back to Learn

How to monitor your brand's AI mentions

Monitoring your brand's AI mentions means asking the same questions shoppers ask, in the same assistants they use, on a schedule you can keep, and writing down what each assistant says about you. There is no single dashboard that shows every AI mention, and no vendor publishes a feed of what assistants say about your brand. The work is a routine: a fixed list of prompts, a fixed list of assistants, a fixed log, and a regular review of what changed. A mention you catch early is a wrong fact you can fix at the source before it spreads. A mention you never check is a wrong fact that keeps answering shoppers while you assume everything is fine.

Start by deciding what counts as a mention. A mention is any answer that names your brand, cites your site, or describes your products, whether the description is right or wrong. Count a citation to your page, a naming of your brand in a list, and a description of your products as three different kinds of mention, because they fail in different ways. A page can be cited with the wrong price, a brand can be named with the wrong products, and a product can be described with a competitor's features. Your log needs a column for each kind, or you will count a wrong mention as a good one.

Where it goes wrong

Monitoring fails quietly, because the routine looks like it is running while it measures the wrong thing.

  • The prompt list is too narrow. The team tests only the brand name, and never tests the category questions shoppers actually ask, such as the best product for a use, the cheapest option, or the closest store. The brand can be invisible in category answers while looking fine in brand name answers.
  • The wording changes every time. Each test uses a different phrasing, so the log cannot show whether an answer changed or the question changed. A monitoring routine needs the same wording every time, or it measures noise.
  • Only one assistant is tested. The brand looks fine in ChatGPT and is wrong in Perplexity, or fine in Perplexity and missing from Gemini. Each assistant reads different sources, so one assistant is not a proxy for the others.
  • Citations are not saved. The team notes that the answer was wrong, but not which page it cited. Without the citation, the next step is a guess, and the fix goes to the wrong page.
  • The log has no baseline. The first test is treated as the normal state, and a slow decline looks like nothing changed. A baseline taken when the site is in good shape is what makes a later change visible.
  • Training answers and search answers are mixed in one column. A training based answer that cannot move yet is read as a failed fix, and a search based answer that moved is missed. The two types need separate columns, because they move on different paths.
  • The review never happens. The log fills up, nobody reads it, and the wrong mention sits for months. A routine without a review is storage, not monitoring.

The steps

Step 1. Build the prompt list shoppers actually use. Write ten to twenty prompts: your brand name alone, your brand name plus your main product, your main product category without your name, a comparison question, a price question, and a local question if you have a store. Use the words shoppers use, not the words your team uses internally. Keep the list fixed for the next round of tests, and change it only when you add a product or retire one.

Step 2. Pick the assistants and the schedule. Test ChatGPT, Perplexity, Gemini, and Claude, since each reads different sources and each has its own search crawler or index. Pick a schedule you can keep: once a week for a small catalog, more often around a launch or a price change. Put the test on a calendar, with one named owner. A routine with no owner stops the first time the owner is busy.

Step 3. Run the prompts the same way every time. Ask each prompt with the same wording, in a fresh conversation where the assistant has no memory of your earlier tests. Save the full answer, every citation or link, and the assistant that gave it. Note the date you ran the test in your log, not in the article or the page, since the log is your record and the page is your public face. Do not edit the prompt mid round to get a better answer. A changed prompt is a new test, not a retest.

Step 4. Score each answer in the log. For each answer, record four things: whether your brand was named, whether your page was cited, whether the facts were right, and whether a competitor was named instead. Use plain words for the score, such as named and correct, named but wrong, cited but wrong price, or missing entirely. Add a note for the exact wrong sentence when the facts are wrong. A log with these four columns turns a pile of answers into a picture you can act on.

Step 5. Trace every wrong mention to its source page. Open each citation behind a wrong answer and note the page that carries the old fact. Check whether the source is your own site, a retailer, a directory, or a review site. Fix your own pages first, then ask other site owners to fix theirs, following the correction guide for wrong answers. Do not report the answer before the source is fixed, since a report about a fact that is still wrong on the web changes little.

Step 6. Watch the traffic the assistants send you. Check your analytics for visits that arrive with a source marker naming an assistant, and check your server logs for the search crawlers: OAI-SearchBot, PerplexityBot, and Googlebot. A visit from a crawler that returns a success response is evidence your pages can be read. A run of errors is a warning that the next round of answers will go stale. Keep the log of crawler visits next to the log of answers, so a missing mention and a blocked crawler can be connected.

Step 7. Review the log on a fixed rhythm and act on the change. Once a month, read the log as a whole. Look for a brand that is named less often, a product that stopped being cited, a price that keeps coming back wrong, or a competitor that keeps winning the same question. Each pattern has a different fix: a missing brand is usually access, a wrong price is usually a source page, and a lost question is usually a better competitor page. Pick the largest pattern, fix it, and note the fix in the log so the next review can tell whether it worked.

The machine-readable piece

This log format is your own tracking sheet, not a vendor format. No vendor defines how you should log AI mentions, so this file is a working template you can copy into a spreadsheet. The columns are the four scores from step 4, plus the citation that lets you trace a wrong answer. Keep one row per prompt per assistant per test run.

prompt,assistant,brand_named,page_cited,facts_right,competitor_named,citation_url,notes
"best waterproof hiking boots",ChatGPT,yes,yes,yes,no,https://[yourdomain.com]/products/example-boots,"correct price and sizes"
"best waterproof hiking boots",Perplexity,yes,no,no,yes,https://[retailer.example]/example-boots,"old price cited from retailer"
"[yourbrand] hiking boots",Gemini,yes,yes,no,no,https://[yourdomain.com]/products/example-boots,"wrong sizes listed"

This robots.txt snippet keeps the search crawlers open so your pages can be re-fetched between test rounds. OpenAI documents OAI-SearchBot and GPTBot as independent settings, and Perplexity documents PerplexityBot as its search crawler. If a round of tests shows your brand missing everywhere at once, check this file and your server logs before you change any page.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Googlebot
Allow: /

What it looks like when it works

On the page, nothing looks different to a human reader. The change is in the routine: the same prompts run on the same schedule, the log fills up with saved answers and citations, and the monthly review turns the log into fixes. A wrong price is caught in the first round after it appears, traced to its source page, and fixed before it answers many shoppers. A missing citation is caught the same way, and the access check or the page fix follows within days, not months. The log shows the brand named more often over time, cited more often, and correct more often, and each improvement can be tied to a fix you made.

Say you run an outdoor gear store selling hiking boots, tents, and rain jackets. You build a list of fifteen prompts, including your brand name, your best boots, and the category question for waterproof hiking boots. You test ChatGPT, Perplexity, and Gemini once a week, and you log each answer with its citations. In the third round, Perplexity names your boots but cites a retailer page with last season's price. You trace the citation, ask the retailer to update its listing, and check that PerplexityBot can fetch your own product page. In the next round, Perplexity cites your page with the current price. The log shows the fix worked, and the monthly review shows the pattern: retailer listings are your most common source of stale prices, so you add a retailer check to every price change from now on.

The fundamentals

These mechanics are shared across every engine and covered in full in their own guides:

What this guide does not cover

This guide covers monitoring your brand's AI mentions with a fixed prompt list, a fixed log, and a regular review. It does not cover correcting a wrong answer, which is a different job with its own guide, or removing outdated product info, which has its own guide too. It does not cover the full technical audit that checks every crawler and every page, or the transaction layer that lets an agent complete a purchase. It does not cover social listening for what people say about your brand on social networks, which is a different kind of monitoring with different tools. Those are separate jobs, and this guide is the routine, not the whole program.

Check your own store now

Run the free scan. One click, no signup, and it shows whether AI crawlers can read your store today: aizii.ai/aeo-scan

Run the free scan

Sources

Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.