AI visibility tracking: why DIY takes 32 hours a month
AI visibility tracking
generative engine optimisation
ChatGPT citations
AI search monitoring
answer engine optimisation
brand mentions in AI
GEO tools

AI visibility tracking: why DIY takes 32 hours a month

Checking if ChatGPT recommends you takes minutes. Doing it properly, across engines and weeks, takes about 32 hours a month by hand. Here's the breakdown.

Nigel Jay Cooper

Nigel Jay Cooper

Co-Founder & Chief Storyteller. Literary fiction author and product builder.

15 November 2026
LinkedIn Profile

Checking whether AI like ChatGPT recommends you takes a few minutes. Doing it properly, across engines, wordings, questions and weeks, takes about 32 hours a month if you do it manually. That's the part to hand to a tool.

Why checking by hand gives you a snapshot

Open ChatGPT, ask "who should I use for [your service]?", and see if you're named. Most people start there, and it's a fair first look. It tells you what one engine said, in one wording, on one day.

It covers very little. A University of St. Gallen study (Schulte, Bleeker & Kaufmann, arXiv:2604.07585) found that only 34–42% of cited sources overlap between consecutive days. Separately, Kevin Indig's "The Consensus Gap" in Growth Memo (11 May 2026) found that just 2.37% of cited URLs appear across all three engines for the same prompt.

Put those together. Yesterday's answer and today's share barely a third of their sources, and ChatGPT, Gemini and Perplexity almost never agree with each other.

If you're named once, you can't tell whether you're recommended or lucky. If you're missing once, you can't tell whether you're invisible or unlucky. Measuring this properly means asking repeatedly, across engines and wordings, and keeping the results in one place.

What doing the measuring in-house costs

Take a worked example. It's arithmetic on made-up numbers, not a client case.

Say you track 20 customer questions. You ask each in three phrasings across four engines, which gives you 240 answers per round. Re-check weekly and that's 960 answers a month.

Reading each one to see who's named and which sources are cited takes, say, two minutes. That comes to 1,920 minutes, or 32 hours a month.

For a marketing lead in a 10–50 person firm, that's most of a working week spent on the spreadsheet instead of the work the spreadsheet is meant to inform. It also leaves out the hardest part, which is deciding what to publish once you've seen the gaps.

How we measure our own questions

We've been running our monitor on our own tracked questions since July 2026. That's internal testing, not third-party research, and it's logged 1,830 samples across seven engine lanes so far.

The design follows from the overlap numbers above:

  • Every paid plan asks each tracked question in three different phrasings, rotated weekly, so a result isn't the luck of one wording.
  • The monitor checks the least-recently-sampled questions first, so the whole set is covered across the month and the same few questions don't get all the attention.
  • Starter re-checks monthly. Basic and Growth re-check core questions weekly, with the wider set, site crawl and digest monthly.
  • When we get cited for a question, the monitor proposes several adjacent questions to track next. Nothing is added until a person approves it.

A person could build every one of those habits into a spreadsheet. Most don't keep them up for months, which is the point at which a tool starts paying for itself.

What has to stay in-house

Tracking tells you where you're missing. It doesn't tell AI why you're worth recommending, and that comes from your people.

A University of Toronto study of consumer-electronics queries (300 queries, four engines; Chen, Wang, Chen & Koudas, arXiv:2601.16858, January 2026) found that the sources AI cites depend on what the asker is trying to do. For consideration queries like "who should I use for X", 59–86% of cited sources were third-party. For transactional queries, 52–68% were brand-owned.

Your own pages need to answer the buying questions, and the choosing questions are decided largely by what others say about you. Both take work that only someone close to the business can do.

The content side has its own evidence. A Princeton and IIT Delhi paper presented at KDD 2024 (Aggarwal et al., arXiv:2311.09735) tested ways of changing content and found that adding statistics raised visibility by about 33%, adding quotations by about 41% and citing sources by about 28%. Keyword stuffing lowered it by about 8%.

The first three all need material that already exists in your business: your own figures, a real quote from the person who did the work, a source you trust. No tool has your figures or your team's quotes.

That's why our drafts are written in each team member's own voice, and why a team member's claim can't be used in a draft until they've approved the wording themselves.

The expertise has to come from the expert. The tool's job is to make it easy for them to hand it over and get it published.

A split you can start this week

You don't need to decide between tool and in-house today. Run the in-house version for two weeks, and the numbers will tell you which parts to hand over.

  1. Write down ten questions a customer asks before choosing a business like yours. Use their words, not your service names.
  2. Ask each question in three phrasings across at least two engines. Record who's named and which sources are cited.
  3. Repeat it a week later. Count how many answers changed. That tells you how much one check is worth.
  4. Give each question an owner among the people who know the answer best. A question with no expert behind it will produce beige content, and AI has no reason to quote beige.
  5. Time the tracking. When it takes longer than the writing, hand it over.

That last step is the one we built our monitor for. It does the repeated asking, the rotation and the weekly re-checks, so your marketing lead and your experts spend their hours on what only they can write.

If you'd like to see where your business is named, and where it's missing, start at ghostart.io.

Connect with me on LinkedIn | Ghostart Platform

Is AI recommending your business?

When people ask ChatGPT, Claude or Google's AI who to use, most businesses never come up — even ones that rank well on Google. Ghostart finds the questions you're missing from, then helps you win them with content that sounds like you, not like everyone else.