init.news●A FIELD JOURNAL
FOR HUMAN OPERATORS
AI discovery / FIELD GUIDE10 MIN READ

SEO, AEO, GEO: what actually changes in the build?

An owner-friendly checklist for search and AI discovery: what to fix, what to measure, and which new platform features actually apply to your business.

Get the useful answer ↓

THE USEFUL ANSWER

SEO, AEO, and GEO describe overlapping discovery goals. For implementation, start with accessible, indexable pages that answer a real question and make their evidence easy to inspect.

Add accurate metadata and internal links, then measure search visibility, observed citations, and useful visits separately. A 2 to 9 November review of inspected first-party search guidance did not establish a new build layer or special AI markup requirement.

None of these steps guarantees an AI citation.

A beam of blue light reveals an ivory page inside a large circular concrete aperture.
Make the answer visible before optimizing how people and systems find it.AI-GENERATED EDITORIAL ART

WHAT YOU'LL LEAVE WITH

A release checklist and one-page discovery triage that distinguish access, indexing eligibility, discovery signals and measurable outcomes, plus a URL-change rule and control map for owners.

  • Use hyphens for word boundaries in new URLs, but do not rename a working live page without a migration reason and redirect plan.
  • Choose the control that matches the outcome: privacy, indexing, snippets, or participation in Google Search generative AI are different decisions.
  • Treat crawler access, indexing, citations and useful visits as separate evidence; none proves the next one happened.
Publish a useful answer, then observe its discoveryA proposed sequence connects useful content, a crawlable page and a repeatable observation. The model contains no ranking, indexing or citation result./ AI DISCOVERYA WORKING MODELILLUSTRATION / NOT MEASURED RESULTSUSEFUL ANSWERCRAWLABLE PAGEOBSERVEKEEP EACH METRIC DISTINCT
Publish a useful answer, then observe its discoveryILLUSTRATIVE WORKFLOW

01One page, several discovery paths

SEO usually describes discovery through search results. AEO emphasizes direct answers. GEO emphasizes visibility within generated responses. These labels are useful when discussing goals, but they do not define three independent technical stacks. Our recommendation is to give one useful page a clear job before buying three kinds of optimization.

Google says established SEO practices remain relevant to AI Overviews and AI Mode. Indexing and snippet eligibility matter, and neither guarantees inclusion. Its newer optimization guide also points to the Search Console control for inclusion in generative AI features. Check the current setting and any inherited choice before assuming an accessible page is eligible. [1] [4] [5]

02This week's useful result is a delta check, not a new acronym

For 2 to 9 November 2026, we searched inspected first-party guidance from Google, Bing, OpenAI and Anthropic for a material change to crawling, indexing, structured data or answer-engine eligibility. The review did not establish a new implementation requirement.

Google announced a Search Central Live event, and Anthropic's current crawler guide remained useful for provider-specific policy, but neither is evidence that every site needs a new file, schema type or publishing system. This is a bounded source review, not proof that no platform changed anywhere. [1] [4]

Swipe or scroll to compare

Weekly checkChange the build only when
Official announcementA named platform documents an applicable behavior, eligibility rule or control.
Current technical guideThe deployed page conflicts with a requirement or supported recommendation.
Owned observationA reproducible access, indexing, answer or action failure identifies the limiting step.
Third-party claimIts primary source and affected decision can be verified. Otherwise keep it as a lead.

A quiet week is not a reason to rename ordinary publishing work. Keep the existing page stable, refresh the source date where the guidance was actually rechecked and spend the next change on an observed reader decision or technical mismatch.

03Build in this order

Swipe or scroll to compare

LayerShipVerify
AccessUseful content in the returned HTML; a stable URLA logged-out visitor can read the page without executing application code
MeaningA descriptive title and a direct answer followed by reasoningA reader can tell which problem the page resolves
EvidenceDated sources, explicit limitations, reproducible artifacts where availableEach factual claim has support; interpretation is labelled
RelationshipsRelevant topic links and a small number of useful next readsThe article can be reached from another indexable page
MetadataCanonical URL and accurate Article dataStructured data describes the visible page
FeedbackA query set and a dated release recordVisibility and useful visits are compared over a stated period

Article structured data can describe the author, publication date, headline, and other applicable properties. Google recommends validating markup and checking the deployed page. Treat this as a description of the article, not a mechanism that certifies its truth or promises selection. [2]

04Before you buy another visibility tool

If you are doing sales, delivery, and marketing yourself, start with one page that helps a suitable customer make a decision. Use the table to choose the next job. These are our suggested priorities, not measured promises of more traffic or sales.

Swipe or scroll to compare

What you seeNext useful actionWhat to keep as evidence
A customer cannot tell whether the offer fitsState who it serves, the actual scope, constraints, and the next step.The customer question and the exact passage you improved.
The page is unavailable or hard to findCheck public access, indexing directives, canonical URL, and a normal internal link.The returned page and the checks performed.
You want visibility in Google AI answersInspect the Search generative AI setting and its inheritance in an authorized Search Console property. [5]The effective setting; write unknown if you cannot inspect it.
A supplier proposes special AI files or tiny keyword pagesAsk which documented requirement the work satisfies. Google says llms.txt does not affect its visibility or rankings and does not require fixed-size chunks. [4]A named platform requirement and a concrete reader benefit.
You do not know whether people see the pageKeep a small query panel; use the dedicated Google report when available.Observed citations, owned impressions, and useful visits in separate records.

Google’s inclusion control is under Settings → Search generative AI. Include is the default for properties without an inherited choice; a child can inherit its parent’s setting. The control concerns participation in specified Search AI features and is separate from AI training controls. The help page says worldwide rollout completed on 31 August 2026. This is established guidance rechecked today, not an October launch. [5]

05A cleaner URL is useful. A reflexive migration is not.

Google’s 23 October documentation note clarifies that hyphens mark word boundaries more clearly than underscores in crawlable URLs. The detailed guide recommends descriptive words, the audience’s language, as few unnecessary parameters as possible, consistent case, and hyphens when you create a URL. This is documentation guidance, not evidence that replacing one character on an established page produces a ranking gain. [14] [15]

Swipe or scroll to compare

SituationDecisionEvidence to keep
You are creating a new pageUse a short descriptive path with hyphens between words.Chosen canonical, title, sitemap entry and internal link.
A working URL already uses underscoresLeave it unless a broader usability, information-architecture or migration reason justifies the change.Current traffic/index evidence, migration reason and expected maintenance cost.
A URL genuinely must changeMap old to new with a direct redirect; update canonical, sitemap and internal links together.Redirect test, final status, destination canonical and rollback owner.
Filters or tracking create many variantsRemove unnecessary parameters and investigate crawlable duplicates before adding more pages.Representative variants, response behavior and the intended canonical inventory.

06Choose the control by the outcome you need

Google’s 21 October update brings existing controls into one current guide. It does not introduce a new ranking lever. The practical improvement is decision clarity: confidential material needs access control; a public page you do not want indexed needs noindex; selected text can use snippet controls; and site-wide participation in specified Google Search generative AI features has its own Search Console control. [14] [16]

Swipe or scroll to compare

Desired outcomeRelevant controlImportant limit
Keep confidential material privateRemove it or require authentication.robots.txt is not access control.
Keep a public page out of Google SearchUse noindex while allowing Google to fetch the directive.A blocked crawler cannot reliably read a page-level noindex.
Limit how text appears in snippetsUse supported robots meta, data-nosnippet or X-Robots-Tag controls.These affect presentation under their documented scope, not every AI product.
Manage inclusion in Google Search generative AI featuresUse the Search generative AI control in an authorized Search Console property.This is distinct from model-training choices and does not promise visibility.

0712–19 October: discovery is not permission, ranking, or a sale

Cloudflare and OpenAI’s 15 October report describes early results from a July research pilot using network metadata to find and refresh pages. The participating cohort covered about 1% of Cloudflare’s network; the authors report hundreds of millions of previously unknown pages indexed. OpenAI still evaluates and ranks pages independently. This is a first-party pilot report, not an inclusion guarantee or a business-outcome study. [8]

The useful lesson is narrower than “move your website.” The report identifies nonexistent URLs, catch-all homepage redirects and duplicate content as wasted discovery work. Its signals do not grant access permission, and this research concerns search metadata rather than sharing page content for model training. Our recommendation: inspect one important URL and its failure cases before changing infrastructure. [8]

On 16 October, Cloudflare described continuous evaluation of Verified bot behavior, including respect for robots.txt and responses to HTTP 429. The system is being used internally; public trust scores and customer-selectable thresholds are future plans in that announcement. Identity alone is not ongoing permission or good behavior. Read “Building trust over time”; do not assume a new customer control is already available. [9]

08Do not build an indexing shortcut for an ineligible page

Google’s 16 October documentation change updates Indexing API quota guidance and removes an outdated form; it is not an announcement that the API now accepts every page. The current default is 200 publish requests per project per day. Its documented scope remains JobPosting pages or BroadcastEvent embedded in VideoObject. An ordinary service page or article does not qualify simply because it needs discovery.

Read the quota page together with the Quickstart before budgeting an integration. [10] [11] [12]

09A 20-minute discovery triage for one business page

This is our proposed worksheet, not a performed customer test. Pick the page people need before buying or contacting you. Save the starting evidence, then choose one action; the time limit is a planning cap, not a measured completion time.

Swipe or scroll to compare

CheckRecordChoose the next action
The real pageFinal URL, response status, readable answer and normal internal link.Fix an observed failure; do not rewrite because one query omitted you.
A deliberately nonexistent URLFinal URL and status; whether it wrongly shows the homepage.Ask the maintainer to return an honest missing-page response, not a successful unrelated page. [13]
The intended discovery routePlatform, content eligibility, access policy and current controls.Use the documented route that fits. Do not relax access restrictions to chase an unspecified benefit.
The customer questionOne question, its answer passage, source and useful next step.Clarify the existing page rather than add a near-duplicate.
The resultEditing/checking minutes, technical evidence, then separately any observed visits or enquiries.Set a review date. Keep business impact unknown until measured.

If the same checks recur, a read-only status-and-link check may be worth automating. Start with a small allowlist of your own public pages, a modest request rate and a human review of exceptions. Compare setup and maintenance effort with doing the check manually. Do not automate policy changes or publishing solely because a citation counter moved.

10Earlier October updates: specialized programs, not shortcuts

On 8 October 2026, Google added documentation for its UGC Fresh Data Program: approved platforms can send fresh user posts and engagement signals through a dedicated pipeline. It is separate from normal web crawling and the Indexing API. Eligibility includes substantial user-generated activity, public stable post URLs, creator profiles, moderation, OAuth 2.0, validated JSON-LD, and timely engagement updates. Participation and appearance are not guaranteed. [6]

Our decision: a business blog or service page should continue with the checklist above. A community-platform owner can assess the program’s eligibility and operating burden before applying. Do not turn company-authored articles into pretend user posts or budget an integration before acceptance. The detailed integration documentation comes after selection. [6]

On 6 October 2026, Search Central published two curated learning paths. Owners can start with Search fundamentals for beginners; developers can use Technical SEO fundamentals for crawl, indexing, and rendering questions. This reorganizes existing learning material rather than announcing a ranking change. Read the topic lists and choose the lesson that addresses your current problem. [7]

11Choose a useful unit of publication

Suppose a studio has built a beautiful, difficult-to-describe interface. A useful publication might explain one design decision, the constraint that caused it, the alternative rejected, and a small artifact another team can inspect. That creates a reason to read and cite the page. Five lightly reworded pages targeting adjacent phrases add less value.

Google's guidance on generative AI content warns that producing many pages without added value may violate its scaled-content policies. It calls for accuracy checks that also cover titles, descriptions, structured data, and image text. [3]

12A small experiment you can run

  • Choose one customer problem and record five natural-language questions people might ask about it. Mark these as proposed queries unless you have observed demand.
  • Publish one substantial answer. Give it one primary search intent and keep related wording on that same page when the answer is substantially the same.
  • Save the publication date and exact changes. Before editing again, inspect whether the page can be fetched and whether its internal links work.
  • Review a consistent query sample after enough time for discovery. Record the engine, date, location, exact prompt, and cited URL. One answer is an observation, not a market-share estimate.
  • When first-party data is available, compare impressions, clicks, and engaged visits. Preserve zero separately from missing data. Keep changes that improve a useful outcome, and document uncertainty.

For your next working session, choose one page and one unresolved customer question. Record time spent on research, editing, and checking. Stop when the page answers that question clearly and the release checks pass. Review discovery later on a fixed date; a quiet result today is not a reason to generate dozens of variants.

13What this does not establish

This is a recommended workflow, not a published uplift study. No ranking improvement or citation gain has been measured for this publication at launch. A search-demand estimate is evidence of possible interest in a query, not proof that the searcher is your customer.

FOLLOW THE EVIDENCE

Sources & checks

  1. Google Search Central: AI features and your website ↗

    [1] Eligibility, query fan-out, standard SEO practices, and no guarantee of inclusion.

    CHECKED 2026-11-09
  2. Google Search Central: Article structured data ↗

    [2] Applicable Article properties, markup validation, and deployed-page checks.

    CHECKED 2026-10-07
  3. Google Search Central: guidance on generative AI content ↗

    [3] Added value, factual review, metadata checks, and scaled-content concerns.

    CHECKED 2026-10-07
  4. Google Search Central: optimizing for generative AI features ↗

    [4] Current guidance on ordinary SEO, inclusion control, llms.txt and content chunking. Existing guidance rechecked; no new October launch claimed.

    CHECKED 2026-11-09
  5. Search Console Help: Search generative AI control ↗

    [5] Effective inclusion choices, inheritance, Settings location, scope and stated 31 August 2026 worldwide rollout.

    CHECKED 2026-10-12
  6. Google Search Central: UGC Fresh Data Program ↗

    [6] Documentation added 8 October 2026, corroborated by the Search Central changelog. Dedicated pipeline, eligibility, acceptance and appearance limits; integration details after selection.

    CHECKED 2026-10-12
  7. Google Search Central: new SEO learning paths, 6 October 2026 ↗

    [7] Two curated playlists of existing material, their intended readers and covered topics; not an algorithm announcement.

    CHECKED 2026-10-12
  8. Cloudflare and OpenAI: making AI discovery better for everyone ↗

    [8] 15 October 2026 first-party pilot report; network metadata, participating cohort, independent ranking, discovery waste and permission boundary. No init.news participation or uplift established.

    CHECKED 2026-10-19
  9. Cloudflare: continuous accountability for Verified bots ↗

    [9] 16 October announcement; behavior signals and current internal use versus planned public/customer availability.

    CHECKED 2026-10-19
  10. Google Search Central: documentation updates ↗

    [10] 16 October entry dates the quota documentation update and removal of an outdated form; not a broad eligibility launch.

    CHECKED 2026-10-19
  11. Google Indexing API: quota ↗

    [11] Current default 200 publish requests per project per day; documentation dated 16 October.

    CHECKED 2026-10-19
  12. Google Indexing API: Quickstart ↗

    [12] Existing JobPosting or BroadcastEvent-in-VideoObject eligibility; rechecked, not presented as new.

    CHECKED 2026-10-19
  13. Google Crawling Infrastructure: HTTP status codes ↗

    [13] Missing-content response behavior; established operational guidance, not a new ranking claim.

    CHECKED 2026-10-19
  14. Google Search Central: latest documentation updates ↗

    [14] Official dates and scope for the 21 October content-control consolidation and 23 October URL word-separation clarification.

    CHECKED 2026-10-26
  15. Google Search Central: URL structure best practices ↗

    [15] Descriptive URL structure, hyphen word boundaries, parameter and case guidance, and crawl-efficiency limitations.

    CHECKED 2026-10-26
  16. Google Search Central: control what you share with Google Search ↗

    [16] Current distinction among access restriction, noindex, snippet controls and the Search generative AI control.

    CHECKED 2026-10-26

Sources stay attached so you can check the argument. Meaningful revisions keep this address.

FROM A GOOD READ TO A USEFUL NEXT STEP

Put one idea
to work.

Pick the problem on your desk. Try the method on the page, or take the whole kit.

The method is public. The kit needs one email address.

FIELD KIT 01 / EDITABLE WORKBOOK / HTML + EXPORTS

The discovery field audit

You have a page that should answer a customer’s question, and no way to tell whether an AI answer can find it or cite it.

Leave with one defensible fix for one page, not a site-wide to-do list.

  • A crawl and page checklist
  • A citation observation log you can repeat
  • An editable audit with CSV export

One page, one sitting. Run it again after your fix.

Get the kitNot available yet

Private downloads are not available yet. No email is collected while this form is unavailable.

The download appears on this page. Nothing is emailed to you, and the optional box stays off unless you tick it.