Why a good intervention changes when it scales
What happened to a grammar checker that worked in trials once many schools began to fit it into their own teaching.
Published November 2026
Institutional detail is generalised. No schools, vendors, officers or products are named, and the checker’s measures and procurement details are left out.
Context
Every student in the system takes English, and marking drafts for basic errors is some of the slowest work an English teacher does. From 2020 we set out to give teachers and students a grammar checker: a tool that finds basic errors in a student’s draft, such as spelling, punctuation and grammar, and suggests how to correct them.
It moved in stages. In 2020 we tried two commercial writing checkers in five schools. In 2021 twelve schools piloted a commercial language checker, and from 2022 it was offered to students from upper primary to upper secondary. Alongside that we worked with a vendor on a grammar checker of our own, trained on local students’ writing and built into a national learning platform. Nine schools trialled it there in early 2023, and from January 2024 it was open to every school on the platform.
A small team drawn from several parts of the system ran it. I worked on it from the first trial in 2020 to the release in 2024, by the end as project lead.
The question
When a tool that worked in a pilot reaches many schools, which parts have to stay the same for it to keep working, and which should schools be free to change?
What made it difficult
The need split by user. Students wanted quick feedback on basic errors, and most tools on the market could give it. Teachers needed to trust that basic errors were caught consistently, so they could spend their marking time on ideas and structure. In 2020 no tool we tried earned that trust.
Accuracy meant three different things. A tool could be measured against the curriculum’s standards, against what a particular teacher expected or against what its vendor claimed. Passing one test said little about the other two.
The checker was only as good as its examples. It had to learn from sentences local students had written, each confirmed as correct or faulty. Schools supplied the marked scripts, and in the trial their teachers did the confirming.
Most school writing was done by hand, and so were the national exams. Typing was new for many younger pupils, and their typing slips showed up as errors.
And the pandemic cut our contact with pilot schools in 2021, the year we most needed to watch the commercial checker in use.
What we noticed
In April 2023 we ran focus groups with teachers from the nine trial schools. I led five of them. Every school had put the checker somewhere different in the writing process.
One school used it at the third draft, after teacher feedback on content and a round of peer review. Another had pupils type a first draft with little planning. Others kept it for journals, speeches or single paragraphs. One used it in place of a writing package. At one primary school, teachers kept it away from the exam year so pupils could practise timed writing by hand.
Teachers disagreed about who it helped. Some said weaker writers gained most, because they could get ideas down without worrying about every sentence. Others said weaker writers couldn’t read the feedback and took it as given, while stronger students asked why. The 2020 trial had shown the same split.
In some schools it added work. Teachers printed scripts to mark by hand, because annotating on screen was harder and the feedback sat beside the text instead of on it. Elsewhere teachers said marking felt easier, though not faster.
Attention moved. Teachers turned to content because they could assume grammar was covered. Some worried that students accepted corrections without reading them. One described writing turning into editing.
The one rule we had fixed was the one teachers argued about. Corrections reached students only after a teacher had checked them. Some teachers wanted stronger students to see them as they wrote. Others found the checker missed errors in weaker writers’ work and wanted to keep checking.
Judgement
Split the need by user. Students’ need for quick feedback on basic errors could be met straight away, so a commercial checker went to students first. Teachers’ need for consistent accuracy could not, so we set out to build a checker for them.
Judge accuracy against the curriculum. Our checker was trained on local students’ writing and checked against the curriculum’s standards, down to the names it gave each error. Accuracy by a vendor’s own definition wasn’t accuracy by ours.
Keep a teacher between the checker and the student. In the platform version, teachers checked the corrections, added what it missed and removed what was wrong before returning the work. Students had to be free to accept or reject a suggestion and to ask a teacher or a friend about it.
Leave the routine to schools. The trial asked each school for at least one piece of continuous writing on the platform and let it decide where that sat in its own scheme of work. We wanted the checker to fit teaching that was already planned.
What changed
We trialled two commercial tools with five schools in 2020 and ran focus groups with teachers and students.
We piloted a commercial language checker with twelve schools in 2021, ran online sessions for pilot teachers and gathered marked scripts to train a checker of our own.
From 2022 students from upper primary to upper secondary could use the commercial checker.
With a vendor, we built a grammar checker trained on local writing and placed it inside the platform as a step in teacher-marked work.
After the 2023 trial, the changes teachers asked for went into the work plan: a less cluttered panel of corrections, filtering by type of error, fewer suggested corrections, plain names for error types and consistent British spelling.
What happened
In the 2023 trial the checker beat the accuracy targets we had set for every type of error we measured. Most teachers said they were comfortable with its accuracy, though they wanted it to do better on harder errors such as phrasal verbs, collocations and run-on sentences. Several were ready to let students use it directly while drafting, which they hadn’t been with the commercial tools.
It opened to every school on the platform in January 2024, still inside teacher-marked work.
Our summary of the trial said teachers were very positive. That was true. The notes held more than the summary: nine schools using one tool in nine ways and no shared view of who it helps.
What I can’t tell you is whether students write better because of it. We couldn’t track how much students used the commercial checker. A longer-term study of impact was planned. I haven’t seen its results, and nothing in my own records follows students’ writing over time.
What transferred
Specify the routine as carefully as the tool.
We were precise about the checker and said little about where it sat in the writing process. Every school filled that gap differently, and the differences changed what the tool did.
Say which parts carry the effect.
Teacher review was the part we fixed and the part teachers argued about. That argument is information, if someone writes it down.
Ask what accuracy means before you measure it.
Measured against the curriculum, against the teacher and against the vendor’s claim, it is three different tests.
A tool that removes one kind of effort moves attention somewhere else.
Teachers moved to ideas. Some students left grammar to the machine. Follow where the effort goes.
Where next
Follow this idea
Operationalise it
Adoption and change in schools (opens in a new tab)Guide
Challenge it
A scalable intervention must specify what may change as clearly as what must remain invariant.Working proposition