Share of AI vs share of search: two measures, two questions

Two measures that answer different questions

Brand teams already know share of search. It is a ratio of how often people search for your brand name compared with the brand names of your rivals, usually taken from a search engine's public trend data. It is a reasonable proxy for how much attention a brand has, because people tend to search for what they are considering.

Share of AI is newer and answers a different question: when a shopper asks an AI assistant a question about your category, how often is your brand in the answer? Someone in Pune asking an assistant for a good protein bar, a gentle face wash for oily skin or a pressure cooker for a family of five is not typing your brand. They are asking for a recommendation, and the assistant decides which names to give.

The two can disagree. A brand with strong recall can be searched by name often and still be missing from assistant answers to unbranded questions, because the assistant has little to draw on about it. A newer brand with well-explained pages can appear in answers before it has much search volume. Reading them together tells you whether your problem is awareness or explanation.

How share of search is usually measured

Pick a set of competing brands. Pull the search interest for each name over a period from a public source. Divide your brand's interest by the total for the set. The number moves with the set you chose, so state the set and the period beside the figure, and keep both fixed when you compare month to month.

Two cautions apply in India. Brand names that are also common words, or that have several spellings in English and in Hindi or other languages, need care, or the figure counts searches that were never about the brand. And search interest is an index, not a count of people, so treat changes as direction rather than as exact size.

How to measure share of AI

There is no public trend feed for assistant answers, so share of AI has to be measured by asking. The method is simple to describe and easy to do badly.

First, write a fixed panel of questions shoppers actually ask in your category, in the wording they use, including questions that do not contain any brand name. Second, ask each question to the assistant, more than once, because answers vary from one asking to the next. Third, save every answer word for word with the time it was given. Fourth, record which brands each answer names and how it describes them. Share of AI is then the share of answers in which your brand appears, out of all the answers in the panel, next to the same figure for rivals.

Each of those choices changes the result. A panel of ten questions gives a very different figure from a panel of a hundred. A branded question always names the brand, so mixing branded and unbranded questions inflates the number. A single asking can flatter or punish you by chance. This is why a share-of-AI figure without its panel, its number of askings and its date cannot be checked by anyone who reads it.

• Report the panel size, the number of askings per question and the dates.
• Report the assistant used. An answer from one assistant says nothing about another.

Keep a record of each answer

For each question in your panel, save the answer, the assistant, the date and the brands it names. Record how the answer was captured. That gives you a checkable basis for calculating the measure described here.

That gives you a share-of-AI reading you can open and question, rather than a score handed over without its rows. It also means we do not publish a share figure from an assistant we are not yet reading. If a supplier shows you a share of AI across many assistants, ask which assistants, which questions and on which dates.

Keep website readiness checks separate from assistant-answer measurements. Confirm that your pages are accessible and their product details are clear, while recognising that this does not guarantee a mention.

Reading the two together

Put the two measures side by side and four situations appear. High on both means shoppers know you and assistants explain you well. High search and low AI suggests the assistant has thin or unclear material about you, which is a content and product-page problem. Low search and high AI is unusual and worth checking against your panel, since a few questions can drive it. Low on both is an awareness problem before it is a content problem.

Whatever you see, avoid two temptations. Do not treat share of AI as market share: it counts mentions, not purchases. And do not promise your leadership that a page change will lift the number. No one can guarantee a mention, so the position to take is to make the pages clearer, ask again after the change and report what was said.

See the related guide on which sources AI assistants cite for where to look when your brand is missing from an answer, and the zero-click AEO scorecard for a way to turn the readings into a monthly review.

A first month, step by step

If you have never measured either, keep the first month small. In the first week, choose three to five rival brands and write the panel of questions. In the second week, pull search interest for the brand names and ask the panel in the assistant, saving everything. In the third week, tabulate the share of search for each brand and the share of answers naming each brand. In the fourth week, write half a page for your leadership that states both measures, the panel, the dates and the assistant, and says what you would need to see next month before drawing a conclusion.

Expect the first figures to be rough. The value of month one is the baseline and the habit of writing the method down. From month two the comparison starts to mean something, because the panel and the rival set stay fixed while the pages, the shelf and the assistant change around them.

Resist reporting the result as a rank. Being named in most answers is not the same as being first among brands, and a league table built from a small panel invites a challenge you cannot answer. Show the raw counts beside any share, so a reader can see how many answers stand behind it.

Finally, agree who owns the next step. A share-of-AI reading that nobody acts on is a report, not a measure. Product page owners, content writers and your agency each hold part of the fix, so send the saved answers to whoever can change the page an answer leaned on.

Common questions