How to Track Your Startup's Visibility in ChatGPT, Perplexity and Google AI Overviews

You can't log in anywhere and see how often ChatGPT recommends your product. There's no rank to check and no single report that covers every assistant. But you can measure enough to see whether things are improving: how often you're named across a fixed set of questions, how you're described, which sources the answers rely on, and how many people arrive from AI assistants.

This guide shows how to set that up as a monthly routine. It covers the prompt set, the tracking sheet, the analytics settings, the webmaster reports and the server logs, and how to turn the numbers into decisions. If you want to know how to get mentioned in the first place, start with our guide to getting your startup mentioned in ChatGPT and Google AI answers. If you've never checked at all, run our one-off AI visibility check first and use it as your baseline. Every product feature and study below was checked on 19 October 2026.

What can you actually measure?

No single number covers AI visibility, so you combine several signals. Each one shows part of the picture and has its own blind spots.

Clicks alone undercount your visibility. In a Pew Research Center study of Google searches, people clicked a link inside an AI summary on just 1% of visits to pages that showed one (Pew Research Center, 22 July 2025 (opens in a new tab)). Many people read the answer and stop, so a mention can reach someone who never visits. That's why the first signal in the table, mentions in answers, is the one this guide spends the most time on.

Step 1: Build a prompt set you'll keep

A prompt set is a fixed list of questions a potential customer might ask an assistant. Running the same list every month is what turns a handful of screenshots into a trend.

Cover the five kinds of question

Buyers ask about a category in different ways, and an assistant can know your product for one kind of question and miss it for another. Include each kind.

Twenty to thirty prompts is a practical size for a startup: enough to cover your main use cases and competitors, small enough to run by hand in an hour or two.

Write prompts the way buyers talk

Take the wording from real sources: questions in sales calls and support emails, comments from your launch, posts in the communities where your buyers gather, and the queries in your Search Console report. People rarely type "top-rated SaaS invoicing platform". They type "what do freelancers use to send invoices".

Version the list

Give each prompt an ID and keep the list fixed. When you want to add or change prompts, save a new version and note the date. Otherwise you won't know whether a change in your numbers came from the assistants or from your list.

Step 2: Decide what to record

For every run of every prompt, record the same fields. A spreadsheet is enough.

From those fields, work out four numbers each month:

• Mention rate: the share of runs in which you're named. If you're named in 18 of 90 runs, your mention rate is 20%.
• Share of voice: your mentions as a share of all mentions of you and your tracked competitors. It shows whether you're gaining ground in your category or just riding a general change in the answers.
• Accuracy rate: the share of your mentions that describe you correctly. An outdated price, a feature you dropped or the wrong audience can cost you as much as being left out.
• Top cited sources: the domains that appear most often in answers for your category. These are the pages assistants rely on, and the places you most want to be listed and described correctly.

Why not track your position in the answer?

Because it doesn't hold still. In a study run by Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai, volunteers ran 2,961 prompts across ChatGPT, Claude and Google's AI features. They found less than a 1-in-100 chance of getting the same list of brands twice, and the same order "less than 0.1% of the time." Their conclusion was that "Visibility percentage across many queries shows more consistency than ranking position" (Search Engine Journal, 30 January 2026 (opens in a new tab)). Gumshoe sells AI tracking tools and the report isn't peer-reviewed, but the practical advice holds: count appearances across many runs.

Step 3: Run it the same way every time

Small changes in how you run the prompts can change the answers, so fix the method and write it down.

• Same day each month. The first working day is easy to remember.
• Same assistants and modes. Pick the assistants your buyers use. For most startups that's ChatGPT, Google (AI Overviews and AI Mode), Perplexity and Copilot. Note whether web search was on, because answers with search can name newer products than answers from the model alone.
• Fresh chats, without personal context. Start a new chat for each prompt, and turn off memory or personalisation where you can, so your past questions about your own product don't shape the answer. Note whether you were logged in.
• Same location and language. Answers can differ by country. If you sell in several markets, run each market as a separate set.
• Several runs per prompt. Run each prompt at least twice, ideally three times, and count each run. With 30 prompts, three runs and four assistants, that's 360 answers a month: too many by hand, which is why many people start with fewer assistants or prompts and add more once the routine works.
• Keep the raw answers. Paste each answer, or save a screenshot, next to your notes. When a number moves, you'll want to read what changed.

Step 4: Set up analytics for AI referrals

Mentions are the main signal, but visits tell you whether being mentioned leads anywhere. Google Analytics 4 makes this easier than it used to.

Use the AI Assistant channel

GA4's default channel group now includes an AI Assistant channel. Google describes it as "the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok" (Google Analytics help, default channel group (opens in a new tab)). Visits are assigned to it when the referrer matches Google's list of AI assistants.

Three things to know before you rely on it:

• It excludes Google's own AI features. Google says the channel "excludes Google's AI Overviews and AI Mode." Those visits count as organic search.
• The list isn't published. Google names examples, not the full list. Check the session source dimension in your Traffic acquisition report for sources such as perplexity.ai or claude.ai, and see which channel they landed in.
• Some AI visits have no referrer. Visits from apps, or from links copied out of an answer, can arrive without one and land in Direct. Treat your AI numbers as a floor, not a total. Links from ChatGPT may also carry a utm_source=chatgpt.com tag, which you'll see in the session source.

Add a custom channel if you need your own definition

If the default channel misses a source you care about, create a custom channel group. Standard GA4 properties can have two custom groups, each with up to 50 channels, and the groups apply to your reports retroactively (Google Analytics help, custom channel groups (opens in a new tab)).

In Admin, open Channel groups under Data display, create a new group from a copy of the default one, and add a channel named something like "AI assistants" with a condition on Source that matches a regular expression. Start from a list like this and adjust it to the sources you actually see:

.*(chatgpt|openai|perplexity|copilot|gemini|claude).*

Order matters. Google's help page says "Traffic is included in the first channel whose definition it matches given the current order of channels in the group," and its own example tells you to move the AI channel above Referral. Do the same, or your AI visits will be counted as ordinary referrals.

Look past the visit count

Once AI visits are grouped, the useful questions are:

• Which pages do they land on? If assistants send people to a blog post rather than your product page, make sure that post has a clear path to the product.
• Do they sign up? Compare sign-up rates for AI visits with other channels. A small channel that converts well deserves more attention than its size suggests.
• Is it growing? Watch the month-on-month trend rather than any single week.

Step 5: Check the webmaster reports

Google and Bing both report on AI features, in very different levels of detail.

Google Search Console

Google says traffic from AI features such as AI Overviews and AI Mode is included in your overall search traffic, in the Performance report under the "Web" search type (Google Search Central, AI features and your website (opens in a new tab)). There's no filter to separate it. What you can still watch:

• Pages you improved for AI answers. If you rewrote a page to describe your product plainly, track that page's impressions and clicks over the following months.
• Branded searches. Searches for your product's name can rise when more people hear of you, including from AI answers they never clicked. A filter on queries containing your brand name gives you a rough awareness trend.

Bing Webmaster Tools, AI Performance

Bing's AI Performance report, in public preview since 10 February 2026, is one of the few direct views of AI citations. It covers "Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations" (Bing Webmaster Blog (opens in a new tab)).

Bing notes that these numbers don't indicate ranking or importance; page-level counts reflect "how often pages are cited, not page importance, ranking, or placement." The report shows citations, not visits, so pair it with your analytics.

Step 6: Check crawler activity in your logs

An assistant can only cite a page its search crawler can reach. Your server logs, or your CDN's bot reports, show whether those crawlers are visiting and which pages they fetch.

Sources: OpenAI's crawler documentation (opens in a new tab) and Perplexity's crawler documentation (opens in a new tab).

What to look for each month:

• Are the search crawlers visiting at all? If OAI-SearchBot or PerplexityBot never appear, check your robots.txt and your CDN's bot settings. Our guide to getting mentioned in AI answers walks through both.
• Which pages do they fetch? If they crawl your blog but never your product or pricing pages, check your internal links.
• Are user-triggered fetches rising? ChatGPT-User and Perplexity-User fetch pages on behalf of a person's request. More of those visits suggests your pages are being pulled into more conversations.

Anyone can fake a user agent, so confirm important visits against the published IP ranges. OpenAI lists OAI-SearchBot's ranges at openai.com/searchbot.json, and Perplexity lists PerplexityBot's at perplexity.com/perplexitybot.json. If you use Cloudflare, its AI Crawl Control (formerly AI Audit) shows which AI services are accessing your content and is "Available on all plans" (Cloudflare docs (opens in a new tab)).

A monthly routine that takes about two hours

The written note matters more than it looks. Six months from now, "mention rate rose from 8% to 15% after we got listed on two review sites" is what tells you what worked.

Turn the numbers into actions

A tracking sheet is only useful if it changes what you do. These patterns come up often, and each points to a different fix.

Competitor mentions are worth a closer look on their own. When a rival starts appearing in answers where it didn't before, it's often after a launch, a new listing or a round of coverage. Our guide to monitoring competitor launches covers how to spot those moves as they happen.

Should you pay for an AI visibility tool?

Paid tools run prompts at a scale you can't by hand, across more assistants, markets and competitors. They're worth it once your manual routine is too slow for the number of prompts you need. Before you pay, ask:

• Which assistants and modes does it check? And does it use the same apps your buyers use, or an API, which can give different answers?
• How many times does it run each prompt? One run per prompt tells you little.
• Does it report mention rate, or a "rank"? In the SparkToro coverage, Rand Fishkin called tools that offer a "ranking position in AI" "full of baloney", and advised: "make sure your provider answers the questions we've surfaced here and shows their math."
• Can you see and export the raw answers? You need them to check accuracy and spot what changed.

A spreadsheet is a fine place to start. It also teaches you what to look for in a tool if you buy one later.

Where Zutto Launch fits

Zutto Launch is our product, so weigh this section with that in mind. We don't track AI visibility, and no launch platform can promise you'll be named in AI answers. What a launch does give you is a public page that describes your product, and one more place on the web where people discuss it.

• A free launch with a product page that stays up after launch day.
• A monthly leaderboard decided by community votes, so you have a month to find your audience rather than a few hours.
• The Weekly Top 3 get a promo kit: a short AI launch video, reels on our YouTube Shorts, Instagram and X accounts, and a section in our newsletter. Votes decide the Top 3; paid placement never buys a spot.
• Monthly winners can earn partner-sponsored rewards.
• Spotlight ($19, one time) adds a 2,500-word story on our blog, a featured listing in your category and a permanent do-follow link with no badge required.
• Premium Launch ($50, one time) adds an AI-generated 30-second launch video, 3 more do-follow backlinks and two posts on our blog comparing you with your alternatives.

Optional paid placement (Spotlight and Premium Launch) is clearly labelled and never changes leaderboard ranking or prize eligibility. See pricing for what each option includes. Zutto opens to waitlist members first.

Your next step

Write your first ten prompts today: two of each kind from the table in step 1. Then run our one-off AI visibility check with them to get a baseline, and put the first day of next month in your calendar for the first monthly run.