Tools for tracking AI visibility

Why monitoring matters

AI visibility is volatile. A brand that surfaces consistently in ChatGPT's answers today may find itself absent next month, not because anything about the brand has changed, but because the model has been updated, the training data has shifted, or a competitor has invested in the citation work that moves them into the answer. Unlike search engine rankings, which change gradually and predictably, AI outputs can shift sharply and without warning.

This volatility makes monitoring essential. Without it, a gallery or artist has no way of knowing whether they are visible in the answers that matter, whether the answers are accurate, or whether their position is improving or deteriorating over time. The cost of not monitoring is not only missed opportunity. It is the slow erosion of a position that was never measured and therefore never defended.

The challenge is that the tools for monitoring AI visibility are less mature than those for traditional SEO. There is no equivalent of Google Search Console for ChatGPT. The platforms do not publish the data that would make measurement straightforward. What exists is a set of emerging tools and practices, each with strengths and limitations, that together can build a reasonably complete picture.

What to measure

The first metric is presence: does the brand appear in the answer at all. This is binary for any given prompt, but over a set of tracked prompts it becomes a rate. A gallery that appears in 30 percent of the prompts where it should appear has a presence rate of 30 percent. That number is the foundation of everything else.

The second metric is accuracy: when the brand does appear, is the information correct. A gallery that is mentioned but described in terms that are outdated, vague, or borrowed from a competitor is not really visible. It is present but misrepresented, which can be worse than absent.

The third metric is position: where in the answer the brand appears. AI answers are not ranked lists, but they do have structure. Being mentioned in the opening paragraph of a ChatGPT response carries more weight than being an afterthought at the end. Tracking position over time reveals whether the brand's standing is improving or declining.

The fourth metric is citation: does the answer include a link back to the brand's site or to a source that mentions the brand. Citations are the bridge between AI visibility and traffic. They are also a signal of the model's confidence: answers that cite sources are generally more stable than those that do not.

The tooling landscape

Several categories of tool have emerged for tracking AI visibility. The first is purpose-built AI visibility platforms, which track a set of prompts across ChatGPT, Gemini, Perplexity, and AI-powered search over time, and report on presence, position, and citation. These tools are new and evolving rapidly, and their coverage varies by platform and market. They are most useful for brands that want a structured, ongoing measurement programme without building their own infrastructure.

The second is manual prompt testing, which is exactly what it sounds like: running a set of representative prompts across the major AI platforms on a regular schedule and recording the results. This is low-cost, high-effort, and surprisingly effective for small programmes. It is also the only way to get a qualitative read on accuracy, which automated tools struggle with.

The third is referral analytics. Google Analytics 4 can be configured to track referrals from chatgpt.com, perplexity.ai, gemini.google.com, and copilot.microsoft.com as a channel group. This does not measure visibility directly, but it measures the downstream effect: how many people are arriving at the site from AI platforms, and what they do when they arrive. It is a lagging indicator, but a useful one.

The fourth is server log analysis. AI crawlers leave traces in server logs, and the pattern of crawling can reveal which pages the platforms are reading, how often, and in what context. This is technical and requires access to raw logs, but it provides a signal that no other source can match: what the models are actually looking at.

Building a monitoring practice

The right approach for most organisations is a combination. Start with manual prompt testing to establish a baseline and understand the qualitative picture. Layer in referral analytics to track the downstream effect. Add a purpose-built tool when the volume of prompts justifies it. Use log analysis as a diagnostic when something changes and the cause is not obvious.

The discipline that matters most is consistency. Run the same prompts on the same schedule across the same platforms, and record the results in a format that allows comparison over time. The value of monitoring is not in any single measurement. It is in the trend, and the trend only becomes visible after several months of consistent data.

For galleries and cultural organisations, the investment required is modest. A monthly prompt test across five to ten representative queries, combined with referral tracking in GA4, is enough to establish whether the organisation is visible in AI answers, whether the visibility is improving, and where the gaps are. That is the foundation on which everything else, from citation building to structured data work, can be measured and justified.