Why does an advisory report usually end up in a drawer?
You get a report with recommendations and stars beside them. Every line is correct, and not one of them says who carries it out, at which address, and in which words.
What exactly do you get in a fix pack?
Three things you can paste and one thing you can check. The FAQs answer the sub-questions that came out of your own measurement; the entity text says in the third person who you are, so an engine can lift it verbatim; the JSON-LD makes both of those machine-readable. The fourth is the jury's verdict, with the score per axis.
What is the difference between generated text and a fix?
That somebody read against it who has an interest in rejecting it. A model assessing its own work approves its own errors more than half as often again, and making it smarter does not help. What does help is a rubric with fixed axes and a jury from another model family — and that the jury does not know who wrote the text. A name attached to a text shifts the verdict in BOTH directions — up or down depending on which name is shown — and it does so even when that name is wrong.
| axis | floor | what the jury looks at |
|---|---|---|
| factual | 70 | Is there anything in it that does not follow from the supplied data? Invented addresses, figures, years or services weigh heaviest here. |
| citable | 60 | Are the key sentences readable on their own and in the third person, so an engine can lift them verbatim without the rest of the page? |
| voice | 55 | Does it sound like your company or like generic marketing language? The voice profile comes from your own texts, not from a template. |
| usable | 60 | Can somebody carry this out tomorrow? Concrete actions with an address, not recommendations in the imperative. |
| compliance | 75 | Unsupported superlatives, figures without a source, or medical, legal and financial promises. The highest floor of the six, and the only one that falls back on sector rules. |
| complete | 60 | Are all the requested parts there, and filled in rather than rushed? |
What does this rest on, beyond our own word?
On the question of what a fix is actually worth, three measurements have an answer that is not ours. Aggarwal et al. compared nine content changes against a baseline across roughly ten thousand questions: statistics, quotes and clear source attributions gave 22 to 41% more visibility in AI answers, and keyword optimisation did less. That is exactly what a fix pack writes — a block of text with a figure, a source and a date. What the study says itself: this is about visibility in a generated answer, not about visitors and not about revenue.
Is this allowed in a regulated sector under professional rules?
That depends on your professional body and we are not a disciplinary authority, so the pack holds to the strictest side. The hard check always runs and without AI: unsupported superlatives and absolute claims get flagged, and in a regulated sector — medical, legal, financial, insurance, property — they get blocked instead of flagged. That difference is not a setting.
Can I not just write this myself with an AI?
The writing, yes. A language model drafts a FAQ in thirty seconds, and it will look polished. What you do not get with it is the question it has to bite on: which question an engine was asked, which engine it was, when, and which address it fetched at the time. Without those four you are writing a good answer to a question nobody asked.
What does a fix pack cost, and what is not in it?
The pack itself is in every paid route; what differs is who applies it. In the watch route you get the measurement and the reason and nothing to paste. In the do-it-yourself route you get the full pack and place it yourself. In the execution route we place it, on the channels you connect, behind a gate you open.
What does ceeme never write for you?
Nothing we cannot derive from your own data. No year you did not give, no certificate you did not show, no headcount we estimated. That is not modesty but the jury's first axis, with the second-highest floor: one invented fact fails the whole pack.
Frequently asked questions
What happens when the jury rejects the pack?
Then it goes back to the writer with the failed axis attached, and that repeats for a few rounds. A few, not as many as needed: a loop without a brake is a bill without a ceiling. If it still does not clear the floor, a human decides — and that is not a fallback but the intended exit.
Is the text unique, or does every client get the same?
Unique, because the input is: the pack is written on the findings from your measurement, with your voice profile taken from your own existing texts. The jury's third axis looks at exactly that — does this sound like your company or like generic marketing — and that axis has a floor of its own.
Can we adjust the pack before we place it?
Yes, and it is recommended. What we do not know is what changes next month, which service you are phasing out, and what is not yet settled internally. The pack is a starting point with the facts already right; the last reading belongs with somebody who knows the company.
Read on
Free, no account and no card.