How PageLens.ai Works: What We Track and How to Read It

PageLens.ai works as a four-step loop. We run buyer prompts across answer engines, trace each answer to the pages it cites, write the missing page for you to approve and publish on your own domain, then re-run the same prompts to see whether the answer changed. An answer engine is a tool such as ChatGPT, Perplexity, Gemini or Google AI Mode that answers in prose and cites sources. A prompt is the question a buyer types into one.

Other products share the name. A cookieless analytics product at pagelens.io and a website-review tool at pagelensai.com are different products. We are PageLens.ai, at pagelens.ai.

We calculate six measures on every tracked prompt set: visibility, recommendation position, citation rate, sentiment, share of voice and a category average. Each has a fixed formula, so a number you see can be checked against the saved answers behind it.

A mention and a citation are different events. An answer can name your brand without linking to your site, so we count them separately.

We save the prompt text, answer engine, timestamp and returned answer for every run. Citation rate does not cover every source a model learned from, and a citation does not prove it caused a recommendation. Use the rates to decide where to publish next, not to claim credit. Our guide to AI brand sentiment analysis goes into run counts for the sentiment label.

We track prompt sets because one answer to one prompt is a sample of one, and answers vary from run to run. In the SparkToro study, the chance of the same brand list in the same order was under 1 in 1,000 (SparkToro). Our research at Citation Research found the same instability, and a second problem: answer engines disagree with each other.

• Repeat runs differ. In Study 001, we asked 100 questions about Indian exam coaching of four answer engines, twice each, on 5 October 2026 from an India-based network location. Only 20% of repeat pairs (80 of 400) returned identical provider lists, and the first-named provider held in 64% of 374 pairs (Finding 3). Our reading is that the first name is steadier than the full list, though it still changed in about a third of those pairs.
• Answer engines cite different places. In Study 007, Google AI Mode cited at least one community platform in 117 of 150 answers, Perplexity in 73, Gemini in 29 and ChatGPT in 3. The study counts community platforms such as Reddit, YouTube, Quora and LinkedIn. Each question was run once per answer engine on 5 October 2026 from India.
• Answer engines disagree on the winner. In Study 001, the four answer engines agreed on the top pick for only 5 of 20 exams.

Both studies are India-based, so the rates will differ for your category. Our inference is that the same instability applies elsewhere, because SparkToro's study found the same pattern in brand recommendations. That is why we do three things: measure many prompts, repeat them on a schedule, and read each answer engine on its own. Our guide to tracking AI citations over time covers the logging side.

As of October 2026, our plans differ by answer engine count, prompt count and how often we read. Launch tracks three answer engines weekly, Growth tracks five daily, and Enterprise tracks seven daily. All plans include unlimited users. Prices are monthly, in US dollars, from our pricing page.

Agency covers multiple client sites in one workspace, with a shared prompt pool and white-label delivery. Growth also adds technical recommendations, user-generated content (UGC, such as posts on forums and community sites) recommendations and Web 2.0 publishing (publishing on third-party sites where anyone can create pages, such as blogging and profile platforms). Enterprise adds the outreach agent, a dedicated account manager and white-label publishing.

The answer allowances equal prompts times answer engines: 100 × 3 = 300 per week on Launch, 100 × 5 = 500 per day on Growth, and 200 × 7 = 1,400 per day on Enterprise. That is one answer per prompt per answer engine per scheduled read.

Dividing price by tracked prompts gives about $2.99 per prompt a month on Launch, $6.99 on Growth and $7.50 on Enterprise ($1,499 ÷ 200 = $7.495). That is our arithmetic. It counts prompt slots only, and the plans also include content.

A change in visibility is real only when it is larger than the sampling margin for the number of answers behind it. At 100 prompts on one answer engine, one weekly read swings by about ±10 points, and comparing two weeks needs a gap of about 14 points. Daily reads shrink that quickly.

This table is our model, built on the standard normal approximation, a textbook formula for the uncertainty in a percentage: margin = 1.96 × √(0.5 × 0.5 ÷ n), where n is the number of answers. The 95% margin is the range within which the true rate falls 95 times in 100. We assume a true visibility of 50% (the widest case), one answer per prompt per answer engine per scheduled read, and independent answers. The margin on a change between two equal periods is about 1.41 times the single-period margin, since two noisy numbers are being compared.

Three conclusions follow, and they are our recommendations. First, a single prompt tells you almost nothing. Read the set. Second, margins are narrower when your true rate is far from 50%, but prompts in one category are related, so the real margin is wider than the table. Treat a move smaller than the two-period margin as unproven, and confirm a larger one over four weeks of the same prompt set before you act. Third, pooling answer engines hides splits. Read each answer engine on its own, as the Study 007 gap between 117 and 3 shows.

Work through eight checks in order, and stop at the first fail. Each has a pass and a fail you can test against the saved answers behind your report.

Visibility is only part of the picture. Pair the report with a look at the visitors answer engines send you; our guide to tracking AI traffic in GA4 shows how to set that up in Google Analytics 4.

Book a demo and we will go through your prompt set, answer engines and approval flow, and show you the plan that fits. Book a demo.

If you want a first read before that, the free AI visibility audit takes your domain, generates buyer prompts for your category and returns a visibility score from 0 to 100, a sentiment split, the competitors named instead of you and the exact words each model used. It takes about 30 seconds, with no signup and no credit card. It shows where you stand once. Tracking weekly or daily is what the plans above are for.