How to Set Up AI Search Visibility Tracking, Step by Step

What AI search visibility tracking measures

Classic rank tracking asks where a URL sits for a keyword. AI search visibility tracking asks a different question: when buyers ask ChatGPT, Perplexity, Gemini, Copilot or Google's AI features about your category, is your brand named, how is it described and is your site cited as a source? There is no position one to hold, only a probability of being included, which shifts by engine, country and run.

So tracking has two halves. The first observes the answers themselves through repeated runs of a fixed prompt panel. The second measures what those answers send you: AI referral sessions, conversions and, where you can capture it, self-reported influence. Setting up both from the start avoids the most common problem teams hit later, which is a visibility chart that nobody can connect to revenue.

Steps one to three: prompts, competitors, engines and markets

Step one: build the prompt panel. Collect 50 to 150 prompts from sales calls, support tickets, onboarding forms and the standard AI patterns: 'best X for Y', 'X vs Y', 'alternatives to X', 'is X legit' and pricing questions. Write them in buyer language with context, and tag each one by funnel stage, product line and market.

Step two: name the competitors. Pick the five to ten brands you actually meet in deals, plus any that dominate AI answers in your category even if sales rarely sees them. Share of voice only means something against a defined set.

Step three: choose engines and markets. Most B2B teams start with ChatGPT, Perplexity, Gemini, Google AI Overviews and Copilot. Add each country and language you sell in as a separate segment, because the same prompt can return different brands in different markets. Keep the list realistic; every added engine and country multiplies the number of runs.

Steps four to seven: runs, storage, analytics and baseline

Step four: schedule repeated runs. Run each prompt several times per cycle on each engine and market, weekly or monthly depending on how actively you are changing things. Repetition turns noisy individual answers into stable rates.

Step five: store full answers. Keep the answer text, the order in which brands appear, the cited URLs, and the date, engine and market. Raw answers allow re-scoring later and settle arguments about what an engine actually said.

Step six: segment AI traffic. ChatGPT search links carry utm_source=chatgpt.com. Create a channel or segment that also catches referrers such as perplexity.ai, gemini.google.com and copilot.microsoft.com, and mark conversions from it. Traffic from Google AI Overviews and AI Mode is reported inside normal Search Console web data, not as a separate source, so treat it as part of organic search. Add an 'AI assistant' option to your 'how did you hear about us' field.

Step seven: record the baseline. Run two to four cycles before launching any optimisation, so you know both the starting level and the normal run-to-run spread.

Worked example: tracking setup for a forex broker

A broker serving the GCC might build a panel of 90 prompts: 30 unbranded discovery prompts ('best regulated forex broker in Qatar', 'which brokers offer Islamic accounts'), 30 comparison prompts against six named competitors, and 30 branded validation prompts ('is Broker X regulated', 'Broker X withdrawal time'). Markets are the UAE, Saudi Arabia and Kuwait, in English and Arabic.

Because the category is regulated, the broker adds an accuracy check on every branded answer: does the engine state the right licence holder, fees and account types? Analytics gets an AI referral channel feeding account-opening events, and the onboarding form gains an 'AI assistant' source option. The baseline then shows something useful within a month: strong coverage in English, weak coverage in Arabic, and one engine repeating an old minimum deposit. That becomes the first work plan.

What a good tracking setup should give you

Check the setup against a few tests. Can you see brand coverage by engine, market and funnel stage separately? Can you compare share of voice with each named competitor? Can you open the raw answer behind any data point? Can you see which pages and third-party sources each engine cites? Can you follow AI referral sessions through to conversions?

If any answer is no, fix it before reporting to leadership. A setup that produces one blended 'AI visibility score' with no drill-down will look fine in a slide and fail the first time someone asks why a number moved.

Agree on ownership as well. Someone has to maintain the prompt panel, read the raw answers each cycle, log site releases and PR coverage, and raise accuracy problems with product or compliance teams. Tracking without an owner tends to decay into a dashboard nobody opens. A named owner and a fixed monthly review, attended by marketing and sales, keep the data tied to decisions.

How Suggesting.ai sets up tracking with you

Suggesting.ai's free 48-hour brand and AI presence audit is the first cycle of this setup: a prompt panel built with your team, competitors defined, five engines covered across your markets, and a baseline of coverage, share of voice, citation share and accuracy. From there we run ongoing independent AI-search tracking with repeated runs and help configure the analytics segment for AI referrals.

Tracking then drives the organic GEO work: entity cleanup, crawler access, answer-ready pages and third-party corroboration, prioritised by the prompts where you are absent. Where your category is eligible, ChatGPT Ads can be measured in the same framework; for financial services outside the US, OpenAI's ad policy generally prohibits them, and the audit confirms eligibility. Tracking is how Suggesting.ai shows whether AI is actually suggesting you more often.