GEO page analysis

GEO-ANALYSIS

URL: https://addition-labs.com/resources/guides/generative-engine-optimization Fetched: 2026-09-11 (raw HTTP + Playwright render via claude-seo runtime; robots.txt, llms.txt, llms-full.txt, sitemap.xml also fetched) Framing: per Google's AI optimization guide, this is SEO applied to AI-search surfaces, not a separate discipline. Scores below are claude-seo heuristics, not Google-internal signals (Google, "Using third-party SEO tools", 2026-06-05).

1. GEO Readiness Score: 77/100

2. Platform Breakdown

3. AI Crawler Access Status

robots.txt (71 bytes) contains only User-agent: * / Allow: / plus the sitemap line. No crawler is named, so every bot inherits the wildcard allow. Each line below is a separate finding.

Search citability (governs whether the page can be cited in that product's search surface)

Training access (governs model-training use only; says nothing about citability)

User-triggered fetchers (ChatGPT-User, Claude-User, Google-Agent, Google-NotebookLM) ignore robots.txt by design; no server-side access control was observed or tested.

No action needed for search visibility. If the owner wants to restrict training use while keeping citability, add explicit Disallow blocks for GPTBot, ClaudeBot, CCBot, Google-Extended and Applebot-Extended; leave the search bots on the wildcard allow.

4. llms.txt Status

Google Search ignores these files (Google AI optimization guide, doc updated 2026-06-29), and no major AI search system has documented consuming them. No citation weight is assigned. The site's own /resources/guides/llms-txt page states the same position, so the files are consistent with the site's editorial stance: published as optionality, not as a lever.

5. Brand Mention Analysis

Wikipedia/Wikidata absence is the largest structural entity gap for ChatGPT-style citation. Google's guide explicitly rejects chasing inauthentic mentions across forums and videos, so the recommendation is limited to accurate entity records (a Wikidata item, a standalone Organization schema) and genuine primary research that others cite, not mention-seeding.

6. Passage-Level Citability

Page body: 3,410 words (parser count), 52 paragraphs, median paragraph ~47 words. The first 30% of the page (roughly the lede through "How nine tested edits performed") carries the core claim, which is where SE Ranking measured ~44% of AI citations originating.

Definition block, first 60 words (strong): > "Generative engine optimization is a named method from a 2024 KDD paper that measured edits to a document already inside an answer engine's context. Its headline gain describes share of an answer, not discovery. A later benchmark found most of those methods ineffective and often harmful to ranking." (48 words, X is ... pattern, self-contained)

Blocks already in the 134-167 word band:

Near-band blocks worth a light edit:

Per Google's guide there is no requirement to chunk content to a word count. The blocks above are reported because they already exist, not as a target to manufacture.

7. Server-Side Rendering Check

Every word of the article, the JSON-LD block, the FAQ <details> elements, the table and the figure captions are in the raw response. The four JS bundles (analyticsEvents, planPicker, legal, page2) inject only navigation and footer. The <h1> sits 278 bytes into <body> with zero links before it, so a crawler with a small read window (the ~5-6K character window measured for OAI-SearchBot) receives the definition and byline before any navigation. This is the strongest technical finding on the page.

8. Top 5 Highest-Impact Changes

1. Name the author and add a Person entity. The byline reads "By the Addition team" and Article.author is an Organization. Google's Who/How/Why test expects a byline where readers expect one; an evidence-heavy research guide is that case. Add a named author with a bio page and Person schema (with sameAs to LinkedIn), keep Organization as publisher.
2. Publish a standalone Organization entity with sameAs. Organization appears only nested inside author/publisher with name and url. Add a root-level Organization (or WebSite + Organization) carrying logo, sameAs: [linkedin.com/company/addition-labs, ...], and the Sheridan WY address already in llms.txt. This is the cheapest entity-clarity gain for AI Mode and ChatGPT, which resolve the brand by entity, not by domain rating.
3. Put the figure data in text. Five of the eight figures are charts (method-ladder, three-of-54, what-bots-ask-and-get, crawl-to-refer, brand-tiers) whose values exist only as pixels; alt text carries the caption, not the numbers. The page's own thesis is that extractable figures help a document get used. A small HTML table under each chart (the nine edit names with their normalised values; the 3-of-54 tally; the Cloudflare shares as read on 3 Sep 2026) makes those claims quotable by a crawler that does not execute JavaScript or read images.
4. Expose the publish date and mark up both dates. datePublished: 2026-07-25 is in schema only; the visible line shows "Updated 10 September 2026" in a plain <span>. Render both with <time datetime="...">. Freshness is one of the stronger AI Mode signals and the page already earns it; make it machine-readable in the body.
5. Declare image dimensions on the seven figure PNGs missing width/height. Lazy-loaded figures without intrinsic size shift layout on load (CLS), which weighs on the classic ranking that feeds AI Overviews. The one screenshot (geo-cseo-bench.png, 1300x590) already has them.

Lower priority, listed for completeness: align <title> ("What the Evidence Supports") with <h1> ("What Survived Testing"); use a page-specific og:image instead of the site-wide og-image.png; RSL licensing is absent (no License: directive in robots.txt, no <link rel="license" type="application/rsl+xml">, /rsl.xml and /.well-known/rsl.xml return 404), add only if the owner wants machine-readable AI-use terms.

9. Schema Recommendations

Detected (one JSON-LD block, 3,744 bytes, valid): Article, BreadcrumbList, FAQPage (4 Question/Answer pairs), nested Organization, WebPage as mainEntityOfPage.

Google's guide says not to over-invest in structured data specifically for AI features. The list above is standard Article/Organization hygiene, which is the part that does count.

10. Content Reformatting Suggestions

Not measured in this run: ranking position (no DataForSEO, no Search Console), Reddit and YouTube mentions, IndexNow status, Core Web Vitals field data. Content-quality heuristic (content_quality.py): overall 80/100, filler 0, AI-pattern matches 0, information density 0.366.