01

AEO — answer engine optimization.

Getting your business named when a customer asks an AI assistant instead of typing into a search box.

Traditional SEO is about ranking on a page of ten blue links. Answer engine optimization is about being one of the three companies an assistant actually names — in ChatGPT, Perplexity, Gemini, and Google's AI Overviews. Some of the industry calls the same work GEO, generative engine optimization. Same discipline, two names; we use AEO.

It does not replace SEO — a site that crawlers cannot read is invisible to both. That is why we score the two separately and never average them.

02

Two scores, never blended.

SEO Health and AI Readiness, each 0–100, each computed from deterministic rules.

SEO Health

0–100

AI Readiness

0–100

Every deducted point traces to a finding row with a URL and a rule ID. There is no composite vanity score to hide behind — two numbers, and the exact reasons behind each one.

03

The competitor table.

Around 25 fixed prompts per business, run against ChatGPT, Perplexity, Gemini, and Google AI Overviews.

For each prompt we record whether you were mentioned, whether you were cited, your position, and — the part that sells itself — which domains got recommended instead of you.

Example — your table is generated from your live results
PromptEngineMentionedCitedCited instead
best plumber in YumaChatGPTnonocompetitor-a.com, competitor-b.com
emergency drain cleaning near mePerplexityyesnocompetitor-c.com

04

How the audit runs.

  1. Collect

    Render the page the way a browser does, and capture both the raw and rendered HTML. Pull robots.txt, the sitemap, llms.txt, headers, and Lighthouse. If any step fails, the crawl fails — no partial-crawl salvage.

  2. Extract

    Normalize every page to rows: title, meta, heading tree, canonical, the JSON-LD graph, links, word count, alt text, Core Web Vitals, and the raw-vs-rendered text delta.

  3. Evaluate

    Deterministic rules, one per file, each returning findings. Every deducted point carries a rule ID and a URL — you can trace it.

  4. Remediate

    Rank fixes by impact and confidence over effort. The model emits the artifact — rewritten meta, a JSON-LD block, an FAQ section, llms.txt, internal links — as a diff, not advice.

  5. Apply

    Approved diffs are written through the site adapter, behind an approval gate. Nothing ships without a human yes.

The whole pipeline runs on the same agent stack CoreShift builds and sells as automation work — which is how one operator delivers it, and why the audit is something we can hand you before you pay anything.

05

What we check on the AI side.

Ingestibility — is the site machine-readable?

Raw-vs-rendered delta
If your content only exists after hydration, most crawlers never see it. This is the highest-signal check we run — and almost nobody runs it.
Bot policy
Explicit robots handling for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and CCBot. We flag accidental blocks and missing policy.
Entity graph
Organization, LocalBusiness, Service, and FAQ nodes with sameAs and @id cross-references — scored as a graph, not as isolated snippets.
Chunkability
Heading hierarchy, answer-first paragraphs, definitional sentences, tables, and stats: what survives retrieval and gets cited.
Attribution
A named author, real credentials, and an accurate dateModified.
llms.txt
Present, and correct.