Recommendations

After each audit (initial or weekly snapshot), zued generates recommendations automatically. They are specific to each URL, chunk, and prompt: not generic advice, but concrete changes tied to the exact analysis that produced them.

​

Four types of output

​

Priority score

Every recommendation has a priority score from 0 to 100. The score is calculated from impact, effort, and how many AI engines are affected. See Prioritization for the full formula. Recommendations are grouped into four categories:

​

How recommendations are grounded

Every page’s recommendations are built on top of a set of observations about that page: what concepts it already covers, how consistently it names its primary entity, which explanatory questions it answers, and which AI engines mention topics it does not. Recommendations are required to cite these observations as evidence, which keeps them specific to the page and hard to confuse with generic advice. Content gaps are prioritized by the strength of the signal behind them. Topics that many AI engines discuss receive the highest priority. Topics another page of your site already covers are surfaced as linking or consolidation opportunities, not as rewrites. Topics a single engine mentions are flagged but ranked lower. Recommendations also favor how the engines actually behaved on your page over general expectations about how a given engine tends to behave. When what we observed an engine doing on your page disagrees with its usual style, and we saw that behavior consistently across more than one response, the recommendation follows what we observed on your page. A signal seen in only a single response is treated as weak and does not override the general guidance. Each new-section recommendation also includes example phrasing grounded in your own entity list, so you have concrete starting sentences rather than an abstract topic brief.

​

When a page cannot win its prompts

Not every page can be cited for every question, no matter what you add to it. For some questions the AI engines consistently reach for a different kind of page than the one you are looking at, and no amount of work on that page changes which kind they reach for. zued checks this before it recommends content work. For each page it looks at what the engines actually cited when they answered that page’s own questions: whether they cited this page, whether they cited you at all, and which of your pages they used instead. Those are counts of what happened, not a prediction. When a page is never cited for its own questions, zued says so above the recommendations rather than leaving you to find out after the work is done. You will see the counts it is based on, and, when one of your other page types does win those questions, which type that is. The suggestion is about where the content belongs, not a promise about what moving it will achieve. This also changes how much you are asked to do. A page that wins nothing gets fewer content recommendations and a plainer explanation, instead of a full list that assumes the page can compete. Technical and structured-data fixes are unaffected: they make a page readable to a crawler, which is worth doing whether or not the page is the one the engines pick. We would rather give you less advice you can trust than more advice you cannot.

​

Tracking progress

A recommendation is something you can act on, so it is something you can close. Mark one done when you have shipped it, or park it when you are not going to, and say why. A parked recommendation stops competing for attention. It leaves the ranked list and keeps your reason with it, so the next person who looks does not have to guess whether it was wrong, blocked, or already true. You can bring it back at any time. The reason matters because the two cases are genuinely different. “We have no capacity right now” is about this month and should come back later. “This is not worth it for us” is about the advice itself and should not. zued treats them differently rather than filing both under dismissed. Recommendations also keep their place between audits. Each one is recognised as the same recommendation when it comes back, even if the wording has changed, so you can see how long it has been open and how many times it has been suggested. Advice that has been asking for attention for five weeks running looks different from advice that appeared this morning. When you implement a recommendation and the next snapshot runs, zued compares the new scores against the previous snapshot, so you can see what moved afterwards.