It finds what is broken, fixes it, and proves it worked.
Most SEO tools hand you a list and leave the work with you. Decifer SEO reads every page, decides by rule what is wrong, writes the change, ships it through your own tests, then goes back four weeks later and tells you whether it actually helped.
A list of 300 problems is not help.
An audit that ends in a spreadsheet moves the work sideways. Somebody still has to read 300 findings, decide which eight matter, brief them, chase them, and then never find out whether any of it paid back.
This closes that loop. The checks are rule-based, so a finding is a fact rather than an opinion. The fix is written and shipped. The result is read from Search Console, and of eight realistic outcomes, exactly two are called a win.
What we won't do
- No traffic forecast. Nobody can promise a ranking, so we do not.
- No page published without a person approving it first.
- No fix beyond headlines and descriptions applied automatically — the rest is named and left to a person.
- No verdict where the traffic is too thin to judge. It says so instead.
What the agent does, end to end.
Six stages, and you can stop it at any one of them. The autonomy choice belongs to you, per website, and it can change any time.
It reads the whole site
One box, no form, no tag to install
Type the address and the check starts. A polite crawler that obeys the site's own rules reads up to 500 pages, and you watch it happen in plain English rather than waiting for a report.
What we measure
Every page that exists, what it says, and how fast it loads on a phone.
It decides what is broken by rule
No model gets a vote on this
Speed comes from Google's own test, preferring what real visitors experienced and saying which it used. Readability, mobile usability, accessibility and structured data are each checked against a published standard. The health score is worked out the same way every time.
What we measure
A 0–100 health score, and a task board grouped by what visitors read, speed, the machinery behind the page, and searches to win.
It writes the fix
Aimed at demand, held to the page
A headline or description is rewritten knowing the phrases people in your market really type, but only phrases whose every word already appears on the page. A page that never says Dubai is never handed a Dubai search, however busy.
What we measure
The exact before and after, and the search demand the new wording aims at.
It puts the change live, carefully
Every step can refuse
It re-reads the page as it is right now, finds the one file that produces it, changes only that file, and runs the site's own tests. If they fail, everything goes back exactly as it was. A site with no tests cannot be worked on automatically at all.
What we measure
One readable change, recorded, with a date set to come back and judge it.
It proves whether the fix worked
Or says plainly that it cannot tell
A fix is finished when the search figures answer, not when it goes live. Seven days in, a first look that usually says too early. Twenty-eight days in, the real answer.
What we measure
Click-through rate, judged against the rest of the same site over the same days.
You choose how far it goes
Per website, changeable any time
Watch only, propose and wait for approval, or apply the fixes it is allowed to apply. Everything except headlines and descriptions stays a person's job, and the agent says so rather than attempting it.
What we measure
Which fixes are waiting on you, and how long they have been waiting.
It aims at what people really search for.
The searches you do not rank for are worth more than the ones you do. The agent works out what your site is actually about, asks what people search for around those subjects, and separates the demand you already meet from the demand you never mention.
When a busy phrase is refused because the page cannot honestly stand behind it, it is not thrown away. It becomes the finding that says your site never covers this subject, and a person decides whether it should.
Where the figures come from
Real search volume for the subjects your own site is about, never our estimate. Every volume arrives with twelve months of history, which is read for a season or a direction. Only a peak with quiet months on both sides is called seasonal — growth cannot make that shape.
Question shapes are built from your subjects, then checked against that data. Anything with no figure behind it is dropped rather than guessed at.
When a page is missing, it writes one.
It picks the subject from real search demand, then reads only your own pages for the facts. Any sentence it cannot point at is cut before you ever see it.
It never publishes. Writing a page produces work for a person to approve, and the draft arrives with the demand figure that justified the subject and the pages the facts came from. A rewrite that uses a figure not on the page, makes a claim the page never makes, or describes a different page entirely is refused outright — not corrected and retried. Nothing changes and you are told why.
Three rules make the verdict worth reading.
A tool that grades its own homework is worse than no tool. These are the rules the agent is held to before it is allowed to call anything a win.
Judge the right number
A rewritten headline changes whether people click when they see the page. It does not change how often Google shows it. So the judgement is on click-through rate, never visitor count, because visitor counts move with ranking and the ranking moved for its own reasons.
Rule out luck
A page going from three clicks to five has said nothing. A difference has to be both unlikely to be chance and big enough to matter. Both, not either: on a very busy page a meaningless change passes the first test on its own.
Rule out the tide
The page is compared against the rest of the same site over the same days. If the whole site rose, the page rising with it proves nothing and is not claimed. This is the guard that separates measuring from taking credit for the weather.
Judging a fix needs Google Search Console connected for that site. Without it the agent says it cannot tell rather than guessing. When a fix made things worse the task goes back to be attempted a different way, and the next attempt is told what not to repeat. After the attempt limit you are told plainly that we tried twice and it did not work.
One decision, reasoned end to end.
This is the shape a Decifer SEO decision takes on the command centre: the measured signal, what the agent found, the move, and what it will go back and check. The figures below are illustrative.
Eleven pages describe themselves in words nobody searches.
Signal
Across the site, eleven pages carry a title Google truncates or rewrites, and their click-through rate sits at 1.4% against a site median of 3.1% at the same average position.
What changed
Nothing about the ranking moved. The pages are shown as often as they were; fewer people choose them once shown, which is what a title problem looks like and what a ranking problem does not.
The move
- Rewrite eleven titles and descriptions using only phrases the pages already stand behind, busiest search first.
- Ship each one as a single file change, through the site's own tests, with the before and after recorded.
- Come back at seven days for a first look, and at twenty-eight for the verdict.
Judged on click-through rate against the rest of the site over the same days. If the whole site rises, these pages rising with it is reported as no change, not as a win.
What it reads, and what it writes to.
Read-only, like every connector. Fixes never travel through a connector — they reach your site through the place your site is actually built, and only with your permission.
The crawl, the checks, the health score, the demand figures, the drafts and the verdicts all work today. Applying a change automatically also needs write access to where the site is built, which you grant per site and can withdraw at any time. Asking search engines to re-read a fixed page reaches Bing, Yandex, Naver, Seznam, Yep and Amazon. Google accepts this from nobody, so for Google the only levers are a clean page list and time — and every sentence you read says so.
Point it at your website.
One box, no tag to install, and you watch it read your site live. The first health score takes minutes. See how the engine decides