Insights · curated by AVA + Spark · refreshed weekly

The reference library for Generative Engine Optimization.

The world's brand-relevant questions are being answered by ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews — and most agencies don't know what those answers say. This is the reference library for the new layer. Auto-curated weekly by the same AI brain mesh that powers AVA. Verified by our operator team. Cited by AI. Built so your brand appears in the answers your prospects are actually getting.

Tracked across 6 generative engines

Cited by AI in 12 of 14 sample prompts last week

Updated 2026-04-25

How this page works

This page is the proof of CharmEngine working on itself. Every essay below is drafted by Spark (CharmEngine's narrative writer agent) against our brand canon. Halo (the relevance ranker) selects topics from operator questions in our private preview cohort + the open web. Sentinel (policy guard) checks every draft against our compliance contract before it goes live. AVA (visibility fixer) measures whether ChatGPT, Claude, and Perplexity actually cite the published essays — and if not, generates remediation variants until they do.

Right now, our cited rate across the 6 engines is 87% on agency-tier prompts. We publish that number in real time so you can verify it. Read the methodology →

Featured essay · most cited this week

FIG 01 · GEO citation flow

Foundational

Generative Engine Optimization (GEO): the new SEO layer that actually answers questions.

For two decades, SEO meant winning ten blue links. Now your prospects ask ChatGPT first and never click. This essay defines GEO precisely: the practice of structuring your content so generative engines (ChatGPT, Claude, Perplexity, Gemini, AI Overviews, Copilot) cite you in the answers they synthesize. We cover the citation-flow model, the seven measurable signals, the gap between traditional schema markup and AI-citation eligibility, and the exact prompt patterns AVA uses to detect where you're invisible.

By The CharmEngine team·14 min read·Updated 2026-04-22·Cited by 4/6 engines

Read the full essay

FIG · Citation gap map

AVA in practice

How AVA detects citation gaps across ChatGPT, Claude, and Perplexity

The 50-prompt sampling protocol, the gap-classification taxonomy (missing brand, missing technical authority, weak source, comparison void), and how AVA hands a structured remediation brief to CharmCanvas in under 90 seconds.

11 min read·Cited by 3/6 engines

FIG · Multimodal output

CharmCanvas

Why one remediation brief should produce 8 channel-native variants

Single-format remediations leak 60-80% of their potential reach. We walk through how CharmCanvas takes one AVA brief and produces a 30-second cinematic video, three image variants, an article, schema markup, a podcast clip, a Discord post, and a thread — all under one cost ceiling, all signed into the audit ledger.

9 min read·Cited by 2/6 engines

FIG · Tier-3 isolation

Agency operations

The agency operator's playbook for tier-3 isolation

Why your enterprise prospects ask about Row-Level Security in the first call, and what to show them on screen one. The SQL, the Postgres role split, the per-tenant KMS scope, and the screen-share script that closes a £100k contract.

13 min read·Cited by 4/6 engines

FIG · Cost-ceiling UX

Cost discipline

Cost ceilings before commitment: the £200/mo agency margin reclaim

The audit your CFO actually wants you to run. We map agency AI spend across Zapier, ChatGPT Teams, Notion AI, Otter, Descript, Jasper, and one GEO tool — typically 18% of agency margin. The cost-ceiling pattern that makes bill-shock impossible by architecture.

10 min read·Cited by 3/6 engines

FIG · Audit ledger

Compliance

GDPR Article 25 in practice: what the audit log actually contains

The DPA conversation you're going to have in Q3 starts with what you logged in Q1. A look at what append-only means when a regulator subpoenas it, why our ledger is signed end-to-end, and the SOC-2 Type II prerequisites we ship with from day one.

15 min read·Cited by 2/6 engines

FIG · Founding-50

Prelaunch

Founding-50 invite-list mechanics: what we learned from Superhuman + Arc

Why we kept the queue at 50, how we route applicants by qualifier (the third form field is "How many client tenants do you run"), and what 'pricing locks for life' actually costs us. Honest numbers from the prelaunch curve.

7 min read·Cited by 3/6 engines

Get every new essay in your inbox.

Monthly digest. First Monday. Operator-grade only — no listicles, no AI hype, no "5 ways to use ChatGPT for marketing." Add yourself to the founding-50 cohort and lock pricing for life.

No spam. Email is encrypted at rest, never sold. Unsubscribe in one click.

Methodology · transparency

How we measure AI citation rate.

The 50-prompt sample

Every Monday at 04:00 UTC, AVA runs a 50-prompt sample across six generative engines: ChatGPT (GPT-5.4 + GPT-5.5), Claude (Sonnet 4.6 + Opus 4.6), Perplexity, Gemini (3.1 Flash + 3.0 Flash), Google AI Overviews, and Microsoft Copilot. Prompts are drawn from a stable corpus of 500 agency-operator and brand-marketing intents, randomly sampled with stratification across topic + intent.

What counts as a "citation"

A citation requires: (a) the brand name appears in the engine's answer, (b) the answer attributes a specific factual claim or recommendation to the brand, (c) the attribution is not negative or comparative-against. Soft mentions ("brands like X") count as half a citation; explicit recommendations count as one.

Self-reporting

The "cited by N/6 engines" badge on each essay is updated weekly with the most recent sample. We publish the full sample audit at /insights/citation-audit — including every prompt, every engine response, and every classification.

What we DON'T do

We don't pay for citations. We don't run prompt-injection or jailbreaking. We don't manipulate the engine via specially crafted user prompts. The whole point is to optimize the underlying content + structure so the engines naturally cite us — and we share the playbook so anyone else can do the same.