AI Readiness Index, Q2 2026 results

What this index cannot tell you.

Every index has limits. Most are not published. These are ours, at the same detail we hold them internally. Where a limit is unresolved, it is listed as unresolved.

The headline limit

Markets cannot be ranked against each other.

Separation between markets changes materially with how deeply each one is sourced. That movement is itself the finding. No value in this dataset supports a league table of markets. We publish market dossiers; we do not publish a market ranking.

The reason is a shared input on both sides. Candidate sourcing depth per market correlates with the country prior at r = 0.729, and all five scored signals load positively on that same depth. The methodology page sets out the same figures.

Read the method behind this limit.

Open limits

Six open limits.

Each of these is live in the Q2 2026 edition. None is patched around, and each is stated with what would resolve it.

  1. Recency is approximate.

    Evidence is dated by a coarse, query-bounded proxy, not an exact publication date. A signal from last month and one from last year can carry the same freshness weight. A precise rule exists and has not been applied. Applying it mid-cycle would be a change of method, not a correction.

  2. Some organisations are counted more than once.

    Subsidiaries, country units and group entities under one brand are not fully consolidated. Some groups appear under more than one website. We have merged 16 such groups. 167 more are still being checked, so a group may appear twice. Separate domains stay separate until equivalence is evidenced. Market totals and the published 1,000 may contain parent and group duplication.

  3. No market reached evidence grade A this cycle.

    A market's grade describes the country sources behind its context score. Grade A needs four independent source blocks. We decline sources that repackage Oxford Insights or International Monetary Fund (IMF) data, so the fourth block is not available in most markets.

  4. We do not publish the share of AI-ready organisations in a market.

    We cannot say what share of a market's organisations are AI-ready. Sourcing depth differs by market, so the numerator is not comparable across markets, and no denominator repairs that. It is withdrawn pending a depth-controlled re-source. Sector composition and gate-pass yield are reportable within a market, not across them. We publish no addressable share of a market, and we do not present the screened universe as a market population.

  5. Evidence was collected in English only.

    All 5,820 Q2 records were sourced in English. Japan and South Korea are most affected and both are likely understated. No correction is applied, because correcting two markets alone would introduce a different distortion. The effect is not yet measured, so it is not described as measured.

  6. Absence from the index is not absence of activity.

    Of 5,820 screened, 3,359 cleared the evidence gate and 1,000 are published. A low score can mean the work is invisible, not missing. A candidate that cleared the gate but is not published was scored, and its public evidence was weaker than the published 1,000. That is not a verdict on the organisation.

Planned improvements

These changes are planned for later editions. Until they are made, the limits above stand.

  • Fuller country context for every market, from three types of source
  • A source on innovation and talent in each market
  • A source on digital infrastructure in each market
  • Searches in local languages, not only English
  • Exact publication dates for evidence
  • A sound base for measuring the share of organisations in a market that show AI work
  • Merging parent companies with their subsidiaries and country units

Known bias

What the method favours, and by how much is unknown

Publishing, in English and often

Organisations that publish in English and publish frequently score higher. Local-language sources are not yet included. That is a property of the instrument, not a finding about the organisations.

The 1,000 are a selection, not the whole region

The universe is proxy-led. Sample construction does not attempt a census, and the published 1,000 are the organisations with the strongest published evidence in this quarter's screened universe, not the most AI-ready organisations in Asia Pacific.

Visibility, not verified readiness

The score measures public-evidence visibility. Strategy intent that has not shipped, and budget that has not produced visible output, are not captured.

A market result is not a market verdict

A market's reading describes the public signal its named organisations publish. It says nothing about the market's organisations that the signals do not reach.

Collaboration and scoring

The relationship does not move the score

AIR APAC works with organisations across the region on programmes, advisory and research. Scoring is deterministic and outside-in: the same public evidence produces the same score, whatever the relationship. No organisation can pay to be included, to move up the order, or to be removed, and no relationship with AIR APAC moves a score in either direction.

Found an error in a published entry?

Write to us with the entry and the evidence. We publish every correction. A correction fixes the record. It does not change the method, and a relationship with AIR APAC never moves a score.

Contact us