AI Insights, Topic Insights and Search Insights: How to Read Promptwatch 360 Data in 2026

Key takeaways

• 360 Insights joins three data sets that usually live apart, per page: AI crawler logs, AI-referred visits and citations. It sits under the Analytics tile as AI Insights, Topic Insights and Search Insights.
• AI Insights tells you what happened on each URL. Topic Insights groups pages by theme so you see which subject areas AI rewards. Search Insights overlays Google Search Console and Bing data so you can compare classic search with AI.
• The most common misreading: the LLM Citations column in the Signals table is not the same number as the AI Citations card at the top of the page.
• Microsoft Clarity also has a feature called Topic Insights. It is a different thing (prompt-based, beta, free), so don't mix the two up.
• Read the three pages in order, and start from the pages with the biggest gap between crawls and citations.

What 360 Insights is actually for

Most GEO dashboards answer one question: was my brand mentioned? That's useful, but it leaves you guessing about why. A page might be ignored because bots never reach it, because they read it and don't use it, or because it gets cited and nobody clicks. Those are three different problems with three different fixes.

360 Insights, part of Promptwatch, is built to separate them. It lines up crawler logs, AI-referred visits and citations on a per-URL basis, then gives you three lenses on the result.

Before you open anything, know the setup rules. Each page has a 7-day and 30-day toggle, and the 30-day option stays disabled until 30 days of data exist. On AI Insights and Topic Insights the period ends yesterday, because the numbers come from a daily rollup. Search Insights follows the Google and Bing sync instead. If you just turned on crawler logs, expect 24 to 48 hours before anything shows. The log ingestion docs cover the setup options.

AI Insights: reading the overview cards

The top of the page has five cards. Read them left to right, because they roughly follow the path from discovery to payoff.

• Active Pages is the share of your known pages that saw any activity. If it's over 100%, bots or visitors are hitting URLs your sitemap and Search Console don't know about. That's not an error. It's a to-do: find out what those URLs are.
• AI Crawls counts indexing and ingestion bots such as GPTBot and ClaudeBot.
• AI Citations counts citation-bot requests.
• Citation Clicks counts AI-referred visits that landed on your pages.
• Citation Ratio is the share of crawled pages that also got a citation-bot request or a citation click. Think of it as how much of what models read they actually use.

Citation Ratio is the card I'd watch over time. A site can have healthy crawl volume and still sit at a low ratio, which means models are reading plenty and using little. That points at content, not access.

Below the activity chart are four leader cards: Top AI Crawlers, Top Citation Crawlers, Top AI Referrers and Top Citation Click Countries. They answer who, not what. If one crawler dominates your logs, check the AI crawler insights docs before assuming it's meaningful. Promptwatch Data has tracked how fast crawler mix can shift: Meta-WebIndexer went from about 2% to nearly 38% of tracked AI crawler requests between mid-July and August 9, 2026. A crawler leading your chart this month may not lead it next month. The weekly breakdown in Promptwatch's AI crawler traffic report is a useful benchmark.

Two more things to know. Citation Clicks, Top AI Referrers and Top Citation Click Countries only fill if visitor analytics is installed (see the visitor analytics docs). The page itself works without it, those cards just stay empty.

The Signals table

The table has one row per page and four numeric columns:

• AI Crawls: indexing and ingestion bots. Answer-fetch bots are excluded.
• Citation Crawls: answer-fetch bots such as ChatGPT-User, Perplexity-User and Claude-User, which fetch a page live in order to cite it.
• Citation Clicks: AI-referred visits that landed on the page.
• LLM Citations: how many times the page was cited across your tracked prompts.

Here's the pitfall. The LLM Citations column is not the AI Citations overview card. The card counts citation-bot requests. The column counts citations in tracked prompt responses. A page can have many citation-bot hits and few tracked citations if you aren't tracking prompts that touch it. Both numbers are real, they just measure different events. The citations vs mentions doc is worth five minutes if the distinction is new.

The six signal badges

Each page gets one badge, and the badge is the fastest way to triage.

Untapped is the interesting one. It's the badge that says the page is being read but not used, which is usually a content problem: the answer is buried, the page is thin, or a competitor says it more clearly. Dormant is usually a plumbing problem, so start with robots.txt, the sitemap and internal links before touching the copy.

There's also a secondary Tracking opportunity badge. It marks pages with significant AI referrer traffic but no tracked citations. Humans are arriving from AI answers, and you aren't measuring the prompts that send them. The fix is to add prompts or a monitor for those pages.

You can filter the table by signal, path, AI crawler and citation source, sort it, and export it as CSV. A good first filter: Untapped pages under your highest-value path, sorted by AI Crawls descending.

Topic Insights: from pages to themes

Individual pages are noisy. Topic Insights groups your sitemap pages by theme, so you can ask whether AI likes your pricing content more than your integrations content, rather than reading hundreds of rows.

It needs a project sitemap with discovered or crawled URLs plus crawler log data from the last 30 days. Without both you'll see a setup card instead. The sitemap docs explain how to change which sitemap a project uses. A theme needs at least 3 pages to be ranked.

The explorer is a bubble plot plus a table. Useful controls:

• Sort by Most activity, Most pages, Most crawls, Most citations or Most clicks.
• Switch between Averages and Absolute numbers. Averages are fairer when comparing a 40-page topic with a 4-page one.
• Zoom between All topics, Topic Groups and Subtopics.
• Filter by page type or path.
• Click any topic to open a side sheet with its page types, top pages and subtopics.

Three ranking cards sit on top:

• Best Performing Topics: the themes AI rewards most.
• Worst Performing Topics: themes with at least the median number of pages but below-median activity per page. Promptwatch says to look here first, and I agree. These are the topics where you've already invested in volume and AI isn't responding.
• Most Popular Topics: the themes with the most pages and activity.

A reasonable workflow: open Worst Performing Topics, click the biggest one, read its top pages, and check what page types dominate. If a topic is all blog posts and your category gets cited from product and landing pages, that's a clue. Promptwatch Data found that product pages are the most cited content type in ChatGPT at 22.4% of citations (Sep 28 to Oct 4, 2026), and that landing pages and docs stay cited longest while news loses half its pages within 36 days (pages cited July 6, 2026). So a topic built mostly from news-style posts will decay faster than one built from landing pages and docs.

Search Insights: lining up Google, Bing and AI

Search Insights overlays Google Search Console and Bing Webmaster data on the same pages as AI crawls, citations and clicks. It requires a Google Search Console property or a Bing Webmaster site attached to the project; setup is in the Search Console and Bing Webmaster docs. Per the changelog, it shows clicks, impressions and top queries per page, splits branded from non-branded queries, and lets you open a page to compare its search traffic with AI activity.

The question this view answers best is the one most SEO teams are quietly asking: which pages rank in Google and Bing but never get cited by AI? Those are your highest-leverage rewrites, because the page already has demand and authority. The inverse is useful too. Pages that get cited with almost no search traffic often answer a narrow sub-question, and are worth expanding.

Bing matters more than people think. Promptwatch's own sources-per-response data notes that Copilot citation counts swung widely, and advises making sure Bing indexes and ranks key pages well. See the average sources per response report (page dated Sept 23, 2025): ChatGPT cites about 5 sources per web-search answer, while Google AI Overviews and Perplexity sit around 10.

The branded vs non-branded split is a sanity check. If most of a page's search clicks are branded, ranking isn't driving that traffic, so a gap against AI citations tells you less than it first appears.

Putting the three views together

A simple sequence works well.

1. Start in Topic Insights. Find the worst performing topic with real page volume.
2. Drop into AI Insights, filter by that topic's path, and look at badges. Mostly Dormant means access problems. Mostly Untapped means content problems.
3. Open Search Insights for the same pages. If they rank well in Google and Bing, the content has authority and the rewrite should focus on structure and directness. If they don't rank anywhere, fix the basics first.
4. Check Tracking opportunity badges and add prompts for those pages, so the next cycle has citation data.
5. After you ship changes, come back in a week or two and see whether badges moved.

On rewrites, Promptwatch Data's fan-out research has practical advice: write headings that read like search queries, front-load the entity and category, and build focused pages for each sub-question. The ChatGPT query fanouts report shows measured averages of about 2.15 queries per response in early December 2025 falling to roughly 1.84 by early March 2026, with a return at 1.0 in April. Take the headline numbers with some care, since the page also describes a single prompt triggering "3-8+" searches, which doesn't match its own averages. The format advice holds either way.

If you'd rather have the fixes drafted for you, the Action Board and Content Agent in Promptwatch pick up from this data, and the platform tracks whether published pages get crawled, cited and clicked afterwards. For other options, the GEO software directory lists alternatives.

Don't confuse it with Microsoft Clarity's Topic Insights

Search for "Topic Insights" and you'll also find Microsoft's feature of the same name, announced in the Clarity blog. It's a different tool doing a different job.

Both are legitimate. Clarity's version is a free way to see how AI answers treat a topic and which competitor domains show up beside you. Promptwatch's version tells you how AI systems actually behave on your own site. They complement each other rather than compete.

Common reading mistakes

• Comparing the LLM Citations column to the AI Citations card. Different measurements.
• Treating Active Pages over 100% as a bug. It means unknown URLs are getting traffic.
• Expecting data on day one. Wait 24 to 48 hours after enabling crawler logs, and 30 days for the longer window.
• Judging a topic on absolute numbers when page counts differ wildly. Switch to Averages.
• Reading one week as a trend. Crawler mix and citation behavior shift. Promptwatch Data recorded reddit.com's share of ChatGPT citations dropping from about 4% to 0.5% on August 14, 2026, almost overnight. Sudden swings happen.
• Ignoring Dormant pages that matter. Fix robots.txt, sitemap and internal links before blaming content.

A quick reference

If you want an outside team to run this analysis alongside technical SEO and content work, 1001 SEO Media uses Promptwatch for exactly this kind of GEO reporting. Otherwise, the three pages above are enough to build a weekly routine on your own: topics first, badges second, search overlay third.