Our Agency Stack, 2025 vs 2026: What We Dropped and Kept
An honest accounting of the tools we kept and the tools we dropped moving from 2025 to 2026, and why Promptwatch is the one we run client programs on.
Agencies get asked what is in their stack, and most answers are a list of logos. This is the more honest version: what we kept, what we dropped, and why, moving from 2025 to 2026. We are not going to invent metrics to justify the changes. The reason we dropped a tool was usually that it answered a question we stopped being asked, and the reason we kept one was that it answered a question we now get asked every week.
What we kept
We kept Promptwatch, the platform we run client programs on. In 2025 it was the prompt tracking and citation tool we used to report whether a client was visible in AI answers. In 2026 it is the center of the stack because the questions clients ask have widened. They want to know why they are visible, not just whether. They want to know what traffic the citation drove. They want drafts that a human reviews before they ship. Promptwatch added Agent Analytics crawler logs, visitor analytics with conversions, and Content Agents with a review inbox, and those three are the features that kept it at the center.
The crawler logs are the one that changed the work most. Agent Analytics records the AI crawlers that hit the site, ChatGPTBot, ClaudeBot, PerplexityBot, GoogleOther, and the Meta AI crawler, and traces the crawl-to-citation path so we can see whether a page got cited because it got crawled, or got cited without a crawl we can find, or got crawled a lot and never cited at all. Those are three different problems, and before the logs we treated them as one. Visitor analytics closes the second gap. It uses a lightweight script or a GTM template to track AI-referred traffic and conversions, so a citation is no longer a number we report in isolation. It has a traffic line next to it. Content Agents publish to Webflow and Framer today, with WordPress listed as coming soon, and every draft lands in a review inbox where a human accepts or declines the slot. We keep autopublish off on every client account. The agent drafts. The human publishes.
We kept the classic SEO tooling we use for the traditional side of an audit, crawl and index and rank tracking. That work did not go away. It just stopped being the whole job. We still run it, we just do not pretend it is the GEO answer.
What we dropped
We dropped the standalone "brand mention" alerts that only told us a brand was or was not named in an answer. The information was not wrong, it was just thin. A mention without the crawl context, without the citation source, and without the traffic signal is a data point we cannot act on. Once prompt tracking in Promptwatch gave us the same mention plus the trend, the citation pages, and the crawler path, the standalone alert became redundant. We cancelled it.
We dropped a content production tool that published on autopilot with no review step. It produced volume. It also produced a draft that cited a competitor's claim as the client's, more than once, and the cleanup was worse than the volume was worth. The lesson was not that AI content is bad. The lesson was that unsupervised publishing is bad. We replaced it with the Content Agents review inbox, where every draft waits for a human.
We dropped a dashboard tool that aggregated every metric into one score. The single score was the problem. A client would see the number go up and feel good while the one surface that mattered to them, shopping cards or Reddit citations or a specific prompt, was flat. Aggregation hid the work. We report per surface now.
What we added
We added the MCP connection so the analyst can pull visibility data inside Cursor and Claude instead of exporting it. The hosted server runs at https://server.promptwatch.com/mcp, and we use read tools like getVisibilityTimeSeries, getCitations, getCrawlerTrend, and listContentGapPrompts to build the weekly read without leaving the editor. Read-only keys hide the write tools, which is how we let a client team look at the data without letting them publish. We added Reddit and YouTube citation tracking as a line item on every retainer after the December 2025 social data made offsite impossible to ignore, and Promptwatch citation analytics gives us page-level, domain-level, Reddit, YouTube, and offsite citations with type breakdowns in one place. We added visitor analytics as a required step, not an optional one, so a citation without a traffic signal reads as a partial result. And we added the query fanouts report to the kickoff, so the prompt list we freeze in week one is anchored to how a buyer actually phrases the search, not to a single prompt we guessed at.
The honest part
We did not drop the legacy tools because they were bad. We dropped them because the question moved. A client in 2025 asked "are we mentioned in ChatGPT." A client in 2026 asks "are we cited on the prompts that drive pipeline, why or why not, and what do we publish next to move it." The 2025 tool answered the 2025 question. The 2026 stack answers the 2026 question. The platform we run client programs on is the one that answers the longer question, because it holds the prompt list, the citation trend, the crawler logs, the traffic line, and the draft inbox in one place, and the older tools each held only one of those.
If you want a stack audit on your current setup, and an honest read on which tools are answering the question you are actually being asked, write to us at hello@1001seomedia.com. We will not sell you tools. We will tell you what we kept and what we dropped, and why.