Methodology

2 numbers

The Index reports 2 measures. The Findability Score is live now: a single score from 0 to 100 that combines the 4 measured dimensions below. It is the number you see across the pre-release pool today.

Share of Local Fostering Voice joins the board at launch. It is the proportion of local fostering results that belong to the authority and its hub, not to a private agency. We measure it today. We publish it for every service where we can identify enough of the local results.

The 4 measured dimensions

The Findability Score combines 4 dimensions. Each carries a fixed weight, and the weights sum to 1.00.

We hold further dimensions back by design in this pre-release; they join at the launch lock, when we set the weights for the launch methodology.

Honest data gaps

We can’t fully measure some services. When we can’t reach a service’s organisation root, for example when a firewall blocks automated access, we can’t read the schema dimension. Rather than record a false zero, we suppress the schema score and the Findability Score for that service, and we flag it as an honest data gap. We still show the dimensions we could measure.

In the pre-release pool, we hold 16 of the 147 services this way, 15 authorities and 1 hub. A data gap is not the same as a service that is not yet in the pool. We’re still adding many authorities and hubs and simply haven’t scored them yet, so they carry no score and no gap flag until they enter a pool.

Members we cannot yet measure

The pool is the assessed set: every service with a completed run under this methodology. A data gap, above, is a service we scored but could not measure on 1 dimension. Separate from those, a small number of cohort members sit outside the pool entirely, because we can’t measure their own fostering page.

Today that is 3 services. Hammersmith and Fulham has an own fostering page that is too large to render within our limits. Kingston upon Thames runs recruitment through a shared service and has no own council fostering page to score. Foster East, the East of England hub, serves a site that blocks automated access. We hold these out rather than score a service on a page that is not its own, or record a figure we cannot stand behind. Where a neighbour delivers fostering for an authority, we score the neighbour that runs it, so Cornwall covers the Isles of Scilly.

Our honest promise

Better findability lifts fostering enquiries and grows a public service’s Share of Local Fostering Voice against private agencies. We claim both directly. We do not claim to deliver approvals on our own. Turning enquiries into approved foster carers depends on a service’s own assessment and support, so we treat approvals as a shared outcome of a partnership, not a number the Index delivers by itself.

Sources

Share of Local Fostering Voice, which joins the board at launch, separates the public service from private agencies. We identify those agencies from Ofsted’s Ownership of children’s social care providers in England, as at 31 March 2025. We collect citation and local search evidence from Google and the 4 named AI assistants on each run.

How we measure Share of Local Fostering Voice

This is version slfv-pre.2. Share of Local Fostering Voice measures how much of the local fostering conversation belongs to a council against private agencies. Under the first pre-release definition, every result the classifier did not recognise counted as unattributed, and we withheld a figure until that share fell below one in ten. That threshold treated 3 different things as one: a genuinely unknown domain, an animal rescue that happens to use the word foster, and another authority’s recruitment site all landed in the same bucket, which read as measurement failure when most of it was not.

slfv-pre.2 measures the share of provider voice. The denominator is the results that come from fostering providers: the council and its hub, private agencies and the doorways that route to them, other authorities’ provider content, and the sector’s neutral bodies. We name and report content that is not a provider at all, meaning social platforms, animal welfare, and advice and civic sites, as context rather than folding it into an unknown bucket. The disclosure threshold stays at one in ten, but it now applies only to domains the classifier genuinely cannot place. Register coverage must still reach 75 percent. The AI answer surface remains out of scope at this version.

This is a narrower and more honest denominator, not a lower bar. You can only read a council’s provider-voice share against other providers, which is the comparison a prospective carer meets on the results page.

What we show as context. Some things appear beside your results without entering any score. Paid search adverts are one: where an advert shows on a measured search, we count it and show it as context on your dashboard. Adverts never enter the Findability Score or the Share of Local Fostering Voice denominator, and they can change between runs. Comparison and doorway portals are different: when they rank in the results they already count inside Share of Local Fostering Voice, so they don’t appear as a separate count, and we don’t trace where a portal sends an enquiry.

Versions

Citation denominator (fostering-pre.2), 10 August 2026. 2 of the 5 questions we put to AI assistants ask about fostering in general rather than about a particular area. No service has ever been named in an answer to either, across 2,885 measurements. Under fostering-pre.1, when an assistant failed to answer one of those 2 questions we removed it from the calculation, on the rule that a missing answer tells us nothing. That rule was wrong here, because for these 2 questions we already know the answer: nobody is named. The effect was that a service scored higher when an assistant failed to answer than when it answered, and 58 of 131 scored services carried a score raised by that artefact.

fostering-pre.2 counts those 2 questions whether or not the assistant answered. Scores fell for 58 services and rose for none; the largest fall is 5 points. The 2 questions still run, so if a service is ever named in a general answer it will count. A service can reach at most 60 of 100 on the citation measure, and at most 88 overall, because those 2 questions cannot be won. We are still reviewing those 2 questions, and any change arrives as a new methodology version rather than an edit to a published board.

The August 2026 board was republished as August 2026 (revised) under fostering-pre.2. Both remain in the record. An August figure under fostering-pre.1 and an August figure under fostering-pre.2 are 2 measurements of the same evidence, not a change in performance, and we do not subtract one from the other.

We recorded Share of Local Fostering Voice for the revised August board wherever we could compute it. The original August board carried no such record. Neither board publishes the figure. It joins the public board at launch. This adds no score and changes no rank.

This is the pre-release methodology. We only compare scores within the same methodology version. We’ll publish the launch methodology in full when we lock it. The score carried a longer name until 20 July 2026. The Findability Score is the same measure with the same weights.

The 16 July 2026 baseline was measured across a rolling window between 5 and 16 July 2026. From August 2026, each monthly measurement runs as a single event.

From September 2026 the monthly measurement runs at noon on the 1st of the month. The August 2026 measurement ran on 6 and 7 August. So 25 days separate the August and September boards, not a calendar month. Later boards are a month apart.

August 2026 note: 16 services measured on 6 July 2026 received part of their crawl score from a page later identified as a security check rather than the service’s own site. The August 2026 measurement corrects this. The affected July scores were overstated on that component and are flagged in our records.

From August 2026, an authority or hub is published as a data gap rather than scored when any AI assistant provider fails to answer during its measurement.

The July 2026 Index covered 112 services. From August 2026 coverage extends to 147, adding 35 services measured for the first time.

9 hub services measured on 5 July 2026 lost one AI assistant provider during their measurement, so their July scores drew on 3 of 4 providers. From August 2026 all 4 providers are verified before any measurement runs.