NAIPS — Net AI Performance Score

One number. NPS-shaped. Range −100 to +100. Derived deterministically from the GEOlocus.ai AI Footprint Score (AIFS). Designed for the executive scorecard, not the engineering review.

Where NAIPS sits in our scoring stack

GEOlocus.ai publishes three layered scores. They answer different questions and are read by different audiences.

Score	Range	Audience	Answers
GEO Score	0–100	Engineering	Is the site mechanically prepared for AI ingestion?
AIFS	0–100	Analyst	Composite of SERP signals, internal data quality, and brand authority.
NAIPS	−100 … +100	Executive / Board	Is AI citing us, or is it citing competitors instead of us?

NAIPS is intentionally read like an NPS: positive means net cited, negative means net invisible, zero means coin-flip parity with the rest of the candidate pool. The shape borrows from Reichheld’s Net Promoter Score so that no executive has to learn a new scale.

Formula

NAIPS is a deterministic rescaling of AIFS into the NPS range. There is no separate measurement layer — every AIFS run also produces a NAIPS value, and the two move together.

NAIPS = (AIFS_total − 50) × 2

  AIFS_total  ∈ [0, 100]   (the GEOlocus.ai AI Footprint Score)
  NAIPS       ∈ [−100, +100]

Worked examples:

AIFS 0 ⇒ NAIPS −100 (no AI citation surface detected)
AIFS 25 ⇒ NAIPS −50 (mostly invisible to AI)
AIFS 50 ⇒ NAIPS 0 (parity — cited as often as a random peer)
AIFS 75 ⇒ NAIPS +50 (preferentially cited)
AIFS 100 ⇒ NAIPS +100 (gold standard — cited across all measured surfaces)

Bands

NAIPS bands map directly to the labels GEOlocus.ai uses on the multi-site survey and the comparison-analysis matrix, so the executive scorecard reads the same as the analyst dashboards.

NAIPS range	AIFS equivalent	Label	Plain-English meaning
+67 … +100	83 … 100	Trusted	Regularly cited across major AI surfaces, often without caveats.
+34 … +66	67 … 82	Sometimes Cited	Cited with minimal caveats; partial AI-surface coverage.
0 … +33	50 … 66	Rarely Cited	Cited with multiple caveats; intermittent or low-confidence surface.
−33 … −1	34 … 49	Fragmented	Inconsistent AI presence; site appears for some queries, not others.
−100 … −34	0 … 33	Essentially Invisible	Not cited; AI either has no record of the brand or routes around it.

The cutoffs are deliberately matched to AIFS bands and to the comparison-analysis labels Trusted / Sometimes / Rarely / Invisible used elsewhere in the GEOlocus.ai product. NAIPS does not invent a new partitioning — it relabels the AIFS surface so the executive view stays in lockstep with the engineering view.

Why NPS-shaped

Recognized scale. Every executive reads NPS without instruction. NAIPS inherits that literacy: positive is good, negative is bad, zero is a coin flip.
Asymmetric stakes. The NPS-shape gives invisibility (−) and high citation (+) equal visual weight. A 0–100 score lets a fifty-fifty AI-presence outcome look like a passing grade. NAIPS makes a tie look like a tie.
Quarter-over-quarter shape. Tracking NAIPS delta over time mirrors how revenue teams already track NPS delta. The directional read — getting more cited or less cited — is the only board-level question worth answering.
Independent of pricing. Unlike audit scores tied to a vendor’s product tiers, NAIPS is derived from AIFS, which is derived from observable signals. There is no incentive to mark a site higher than it measures.

SQL view (canonical definition)

NAIPS is implemented as a SQL view over the existing aifs_scores table. The view is the only place the formula and band cutoffs are encoded; product surfaces (comparison analysis, executive dashboards, client PDFs) read from the view, never reproduce the math inline. Below is the exact, current view definition.

CREATE OR REPLACE VIEW naips_scores AS
SELECT
  s.agent_id,
  s.aifs_total,
  s.aifs_band,
  s.aifs_version,
  -- NAIPS = (AIFS - 50) * 2, range [-100, +100], NPS-shaped.
  ROUND(((s.aifs_total - 50) * 2)::numeric, 1)               AS naips_score,
  CASE
    WHEN s.aifs_total >= 83 THEN 'trusted'
    WHEN s.aifs_total >= 67 THEN 'sometimes_cited'
    WHEN s.aifs_total >= 50 THEN 'rarely_cited'
    WHEN s.aifs_total >= 34 THEN 'fragmented'
    ELSE                           'essentially_invisible'
  END                                                         AS naips_band,
  s.scored_at,
  s.updated_at
FROM aifs_scores s;

COMMENT ON VIEW naips_scores IS
  'NAIPS (Net AI Performance Score) -- NPS-shaped rescaling of AIFS into [-100, +100]. '
  'Methodology: https://geolocus.ai/methodology/naips. '
  'Bands match the comparison-analysis labels (trusted / sometimes / rarely / fragmented / invisible).';

GRANT SELECT ON naips_scores TO authenticated, service_role;

Because NAIPS is a view (not a materialized snapshot), every AIFS write — whether from the batch-aifs-score edge function, a manual recompute, or a backfill — flows through to the executive scorecard with no second step. There is exactly one formula, one place to change it, and one place to audit it.

Roadmap — NAIPS v2: true platform-citation probes

The v1 NAIPS published above is a deterministic rescaling of AIFS, and AIFS itself is rooted in SERP signals (knowledge graph, sitelinks, third-party citations) plus internal data quality. That makes v1 reproducible at near-zero marginal cost and aligned with the composite that already powers our analyst dashboards.

NAIPS v2 extends the score with direct citation probes against the four major AI surfaces — Perplexity, OpenAI (ChatGPT), Anthropic (Claude), and Google (Gemini) — using a fixed query set per industry cohort. The probe layer answers the literal NPS question: of the platforms a buyer is likely to ask, how many cite this brand vs. cite a competitor instead?

v2 will land alongside a aifs_platform_probes table (one row per × platform × query × run) and a hybrid view that blends the rescaled AIFS with the platform-citation rate. Bands and ranges will be preserved so the executive scorecard does not need to be re-learned.

Methodology authored by GEOlocus.ai. Published 2026-05-19. See also: /whitepaper, /methodology/sitemap-benchmark, /site-survey.