A weekly AI search report from two fixed tabs: citations first, sentiment second
How to run a weekly AI search report from fixed Citations and Sentiment tabs: what to read first, how to judge week-over-week changes, and what to do next.
Most teams that track AI search end up asking the same two questions every Monday. Which sources do the engines cite when they answer our prompts? And how do those answers describe us next to our competitors? The questions barely change from one week to the next. The dashboard people use to answer them often does, because someone adds a widget, someone else changes a filter, and by March nobody is sure the January numbers were measured the same way.
This guide sets up a weekly report that doesn't drift. It reads citations first, because that is where most fixable problems show up, and sentiment second, as a check on how you're being described. The examples use Promptwatch's Sentiment and Citations dashboard tabs, which shipped on November 13, 2026, but the reading order works in any tool that records sources and sentiment per answer.
Why a fixed layout suits a weekly report
A custom dashboard is the right tool for a question you are asking for the first time. A weekly report is the opposite case: a question you have already decided to ask every week. For that job, flexibility costs you. Every edit to the layout is a small break in the series, and a weekly series is only useful if this week can be read against the last eight.
Promptwatch's changelog describes the new tabs like this: every project has Sentiment and Citations tabs next to Dashboard, each with a fixed layout and saved filters. The layout is locked. A team that wants to change it chooses Duplicate in the tab menu and gets a normal dashboard it can edit. That split is the useful part. The locked tab is the reference view everyone reads, and the duplicate is where an analyst can try things without moving the baseline.
If you do want to build your own from scratch, our guide to dashboard templates covers how to choose widgets.
Part one: the Citations tab
The Citations tab shows your cited domains, the top cited URLs, and cited posts from YouTube, Reddit, and LinkedIn. Read them in that order. Each one answers a narrower question than the one before it.
Cited domains: who the engines trust for your prompts
Start with the domain list and split it into two groups: your own domains and everyone else's. How often your own domain appears is a rough version of what Promptwatch calls the self-citation rate, the share of citations that point at your domain rather than at third-party pages. In Promptwatch's own reading of its metrics, mentions without self-citations point to a content or crawlability problem. So a brand that gets named in answers but rarely shows up in the domain list has something specific to investigate.
Then look at the third-party domains near the top. These are the sites the engines lean on for your topic: review directories, comparison pages, trade publications, sometimes a competitor's docs. The list always moves a little from week to week, so the useful question is narrower. Did a domain enter or leave the top of the list and stay there? A new domain that holds its place for two or three weeks is a source you should know about. Read what it says about you, and check whether you're on it at all.
Top cited URLs: which pages do the work
Domains tell you where. URLs tell you what. On your own site, the top cited URLs usually show a short list of pages carrying most of your citations, and they are rarely the pages the marketing team would have picked.
When one of those pages drops out, check two things before you rewrite a word. Did the page change recently? And can AI crawlers still reach it? Promptwatch's crawler logs (Professional and up on brand plans, every agency plan) show crawler visits and a citation rate per page, which is how you tell an access problem from a content problem. A page that crawlers stopped fetching needs a technical fix. A page they still fetch but no longer cite needs a content look.
For a single URL that matters a lot, like a pricing or comparison page, add it to the Page Tracker so you follow it on its own instead of hunting for it in the weekly list.
Cited social posts: the section most reports skip
The last section lists cited YouTube, Reddit, and LinkedIn posts. This is where engines pick up opinions from people who aren't you, and it behaves differently per engine. Promptwatch's own data is a good reminder to check the wider context before you react. Reddit's share of ChatGPT citations fell from about 4% to about 0.5% on August 14, 2026, in Promptwatch's citation data (our write-up of that drop). A team that saw its Reddit citations fall that week and blamed its own community work would have drawn the wrong conclusion.
So when a social post appears or disappears, first ask whether the change is specific to your prompts or happening across the engine. Then ask whose post it is. A cited thread from your own team is a channel you can keep feeding. A cited thread from an unhappy customer is a support conversation that happens to be public. If you want Instagram in the same view, duplicate the tab and add the Top Instagram posts widget to the copy. Promptwatch added that widget to dashboards on November 4.
Reading changes without fooling yourself
A weekly report is mostly a report of changes, so the comparison period matters more than any single chart. Promptwatch fixed a bug on November 13 that's worth knowing about if you compared periods before then. Monitor comparisons now measure the selected range against a previous period of the same length. Before the fix, Last 14 days was compared with the 15 days before it, which the changelog says skewed changes in visibility, share of voice, and position. A mismatched window is enough to make a flat fortnight look like it moved.
Two practical rules follow. If you kept a log of week-over-week changes from before mid-November, don't splice it onto the new numbers without a note saying where the method changed. And keep your own report's window fixed. Pick a range, use it every week, and compare like with like.
A second fix from the same day affects exports. Exported citation positions now match the Citations table; before, they were one too high. If someone pulls a CSV for a quarterly review and the old and new positions don't line up, that's why.
There's also a speed change that matters more than it sounds. On large projects, the Citations table now loads in under a second, down from about 22 seconds, and Offsite Mentions in under half a second, down from about 6. A weekly review that involves opening ten URLs used to mean a lot of waiting. It doesn't anymore, which makes it more likely someone actually opens them.
Part two: the Sentiment tab
The Sentiment tab holds a sentiment chart and a heatmap of sentiment by brand and topic. Promptwatch scores sentiment up to 100, where about 50 is neutral. Our perception guides cover what the score means, so this section sticks to the practical case: it's Monday, and one cell is redder than it was last week.
When a cell turns red
A heatmap cell is one brand on one topic. A red cell for a competitor is interesting. A red cell for you is a task. Work through it in this order.
- Check how much data sits behind the cell. A topic with only a few prompts can swing on a couple of answers. The Topics page shows how many prompts each topic holds, and Promptwatch's API returns a response count on sentiment cells in its performance heatmap, so you can see how many answers a cell rests on before you escalate.
- Check whether the change is broad or narrow. If your row turned red across several topics at once, look for something that changed everywhere, such as a news story about you. If it's one topic, the cause is more likely in that topic's sources.
- Find the claims. Brand Narrative (every plan) pulls the individual claims AI makes about you out of tracked answers, groups repeats, and puts the most negative claims at the top of its table. Start there.
- Trace each claim to a page. Look at what the answers on that topic cite. If the page is yours and out of date, that's a content refresh. If it's someone else's, it goes to whoever owns outreach. On Business, Agency Scale, and Enterprise, Fact Check does this lookup: it checks the claim against your own pages and the 50 most-cited offsite pages for it, and says whether the verdict rests on your site, on cited sources, or on both.
The Agent Chat panel also opens on the Sentiment page and answers from the charts and filters on screen. "Which answers pulled this topic down this week?" is a question you can put to it instead of digging by hand.
When nothing turned red
Most weeks, nothing will. That's not a reason to skip the tab. Look at the competitor rows instead. A rival whose sentiment drops on a topic you also compete on is an opening, and the same claim-level work tells you what buyers are hearing about them.
Sending it every week
The last piece is getting the report to arrive without anyone building it. In Promptwatch, a scheduled report can follow any of the three tabs (Dashboard, Sentiment, or Citations), and the report is titled with that tab's name. Every brand plan includes 10 scheduled PDF reports. That's enough to send the Citations tab to the SEO lead and the Sentiment tab to whoever owns brand and comms, each on its own schedule.
A setup that holds up over a quarter:
- Set the filters on each tab once, including the date range you chose above, and leave them alone.
- Schedule Citations and Sentiment for the same morning so they get read together.
- Keep a short running note next to the reports with the date, what changed, and what you did about it. The PDF records the numbers. The note records the decisions.
- When someone asks for a new chart, duplicate the tab and build it there. If it's still useful after a month, decide then whether it belongs in the weekly send.
If your team prefers a written summary to a PDF, the weekly GEO briefing with Scheduled Chats sits well next to this routine.
Where Promptwatch fits
You could put this report together from exports and a spreadsheet. The trouble is the second half of each section. Knowing which URL lost its citation only helps if you can then see whether crawlers still reach it, and a red sentiment cell only helps if you can get from it to the claim and the page behind the claim. Promptwatch keeps those steps in one project: the fixed Citations and Sentiment tabs as the weekly front page, crawler logs for the access question, and Brand Narrative for the claims. The tabs and scheduled PDFs are on every brand plan, starting with Essential at $95 a month. Crawler logs start on Professional at $245 a month.
To run this routine on your own prompts, start the 7-day free trial on a monthly plan and schedule both tabs in the first week.