Most SEOs think backlinks affect AI visibility, but few track it. A free four-week log with prompt types, repeat runs and a simple score shows if you appear.
In a 2026 survey of 500 SEO professionals, 74% said backlinks affect AI search visibility, 24% said they track that visibility, and 19% said they had changed how they build links because of it. The numbers come from Reporter Outreach's State of Link Building 2026 report, which describes itself as self reported. They describe a gap between belief and measurement, and the gap has a simple cause: there is no free dashboard for it.
This post is a method for closing the gap on a small scale. It costs nothing but time, and it produces a record you can read yourself, which is more than a belief gives you.
What you are measuring, and what you are not
You are measuring whether AI tools mention or cite your brand or pages when asked a fixed set of questions, and whether the pages they cite are ones where you have links. You are not measuring ranking, and you are not proving that links caused anything. A log shows association. Treat any pattern as a lead for further testing.
Week 0: set the questions
Write twelve questions a buyer in your niche would ask, in the words a person would type, not keywords. Split them into four types, three of each:
- Category: what is the best X for Y. These show whether you are in the AI's shortlist at all.
- Problem: how do I fix X. These show whether your content gets cited for the job you solve.
- Comparison: A versus B for Y, with you as A or B. These show how you are described next to a rival.
- Brand: questions that name you directly. These show what the AI says about you, and whether it is right.
Keep the wording fixed for the whole four weeks, or the results will not be comparable. Real people word the same need in very different ways, so your fixed set is a sample, not a full picture. SparkToro found that across 142 prompts people wrote for the same headphone question, there were barely two that looked alike.
Choose the tools to check: the ones your buyers use. ChatGPT, Perplexity, Gemini and Google's AI Overviews are a reasonable starting set. Use a logged-out or clean browser profile so your own history does not shape the answers, and note the country.
The log
Make one row per run. Record: date, tool, question number, run number, the score, the other domains cited, and a link to each cited page. That is all. Resist adding columns.
Run each question more than once
AI answers change from one run to the next. In SparkToro's study, 600 volunteers ran twelve prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. Rand Fishkin reports a less than one in 100 chance that ChatGPT or Google's AI gives the same list of brands in any two responses, and closer to one in 1,000 for the same list in the same order. So one check per question tells you very little, and your position in a list tells you almost nothing.
What does hold up is how often you appear across many runs. Fishkin concludes that visibility percentage across many prompts run multiple times is a reasonable metric, while any tool that reports a ranking position in AI should not be trusted. Mike Sonders, writing in Search Engine Land, suggests running each key prompt about five times whenever you pull data, since a single spot check is nearly meaningless.
This is our plan for the log, and the scoring below is part of it. Run your category and comparison questions five times each per tool every week, in fresh chats, and the problem and brand questions once. That keeps the work manageable and puts the repeats on the questions closest to a buying decision, the kind Sonders says to repeat.
Score each answer
The scoring is kept simple so you can do it by hand. Give each answer a score of zero, one or two:
- Zero: your brand is not mentioned.
- One: your brand is mentioned, but your site is not cited.
- Two: your brand is mentioned and one of your pages is cited as a source.
For each repeated question, your weekly visibility is the share of runs that scored one or two. For questions you run once, read the four weekly scores together at the end. Ignore where you appear in the list. Sonders groups results into three tiers. Below 20% is the little-seen long tail, 20 to 80% is the visible middle, and 80% or more is the dominant top tier. A question that moves you up a tier across the four weeks is a signal. A single mention is noise.
Weeks 1 to 4: run it, then add the link layer
Run the whole set each week at the same time of day, with the repeats above. After week two, add a second sheet listing every page on other sites where you have a link, with the date it went live. In week four, compare the two sheets: which of your cited or mentioned appearances come from pages that also link to you, and which come from pages that do not?
Our earlier look at brand mentions against backlinks in AI citations found that mentions without links can matter as much as links, so expect some of your appearances to come from pages that never linked to you. That is a finding too.
How to read the result
If your brand appears in few answers and the cited pages are mostly third party roundups, the question is whether you are on those pages. If you appear often and your own pages are cited, your content is doing the work and links may matter less. If you do not appear at all, check the basics first: whether the tools can read your pages and whether your brand has a clear description they can reuse.
Whatever you see, write down what you would change and what you would expect to happen, then check again four weeks later. That turns a log into a test.
What this cannot do
Twelve questions across four tools is a small sample, even with repeats, and AI answers shift as the tools update. This log will not rank you against competitors, and it will not tell you whether links caused an appearance. It tells you whether you are visible for the questions your buyers ask, which is the question the survey says most SEOs have not asked.
If you also want to check the links themselves, SerpCrest's bulk DoFollow checker shows whether each link in your second sheet is live and followed, so the link layer of your log rests on facts.
