GEO Resource Library: Sources, Studies and Paths

This is not a list of every link that mentions "AI SEO." It is a filtered, ranked, opinionated set of sources we actually use, organized so a specific reader — an executive, a marketer, an engineer — can work through it in order and come out knowing more than most of the agencies pitching them.

Every resource below carries a credibility tag. We use four:

• [PRIMARY] — the vendor's own official documentation. Google explaining Google, OpenAI explaining OpenAI. Authoritative about what that vendor's system does, silent about competitors, and never independently audited by anyone outside the vendor.
• [PEER-REVIEWED] — research that went through academic or conference review. Slower to produce, harder to game, and the only category here with a formal replication standard behind it.
• [INDEPENDENT] — analysis from an organization with no commercial stake in whether you buy a GEO service or a GEO tool. Rare in this field, which is itself worth noting.
• [VENDOR] — produced by a company that sells GEO software, GEO services, or both. Useful, often the only data that exists on a given question, and shaped by an incentive to look good in the aggregate.

Here is the honest framing: most GEO content published in 2026 is [VENDOR]. That is not a disqualification. Vendors run the platforms with the retrieval logs, the crawl data, and the citation counts, so they are often the only party with anything to measure. The discipline is to read vendor research for its data and read past its framing — check the sample size, check the date, check what the study does not claim, and treat every headline percentage as a data point rather than a verdict. We apply that same discipline to Artlogic's own claims elsewhere in this hub, including the GEO pillar page.

flowchart TD
    A["New GEO source found"] --> B{"Who published it?"}
    B -->|"Vendor's own official docs"| C["PRIMARY"]
    B -->|"Academic or conference reviewed"| D["PEER-REVIEWED"]
    B -->|"No commercial stake in your decision"| E["INDEPENDENT"]
    B -->|"Sells GEO tools or services"| F["VENDOR"]
    F --> G["Read the data, discount the framing"]

How every resource on this page was tagged: publisher and incentive decide the label, not how confident the source sounds.

Three learning paths

Pick the path that matches your role. Each one mixes Artlogic hub pages with external primary sources, runs in a fixed order, and ends with something you can act on immediately.

The Executive Path (about 90 minutes)

For whoever signs off on funding this. Ends with the questions to ask any agency pitching GEO.

1. "How search actually changed," GEO pillar page — 10 min. Real scale numbers, including Google's roughly 500x referral advantage, before anyone can exaggerate the shift.
2. Google AI features guidance [PRIMARY] — 15 min. Google's written admission that no special file or schema is required — the best filter for a vendor call.
3. Reuters Institute, Trends and Predictions 2026 [INDEPENDENT] — 20 min. The only non-vendor source sizing what you are being asked to fund.
4. GEO paper abstract and conclusion, arXiv:2311.09735 [PEER-REVIEWED] — 10 min. The term's origin and its bounded claim, so no one inflates it into a guarantee.
5. "Common GEO myths," GEO pillar page — 10 min. A rehearsal for the pitches you are about to hear, almost verbatim.
6. GEO Methodology, five-phase overview — 15 min. The shape of a real programme before approving a scope that skips straight to content.
7. "What to be skeptical of," this page — 10 min. Converts directly into the questions below.

Questions to ask any agency pitching GEO:

• Can you show me the robots.txt audit before we talk about content?
• Which of your reported numbers come from your own tool versus an independently verifiable source?
• What is your position on FAQPage schema, and do you know it stopped producing rich results in Google Search on 7 May 2026?
• If you use the word "guarantee" about rankings, citations, or recommendations, stop the meeting.
• What is your measurement method, and how many times do you sample each prompt?

flowchart TD
    A["1. GEO pillar scale-of-shift section - 10 min"] --> B["2. Google AI features guidance - 15 min"]
    B --> C["3. Reuters Institute 2026 report - 20 min"]
    C --> D["4. GEO paper abstract and conclusion - 10 min"]
    D --> E["5. Common GEO myths - 10 min"]
    E --> F["6. Methodology five-phase overview - 15 min"]
    F --> G["7. What to be skeptical of - 10 min"]
    G --> H["Questions to ask any agency"]

The Executive Path: seven steps, roughly ninety minutes, ending in a question checklist rather than a decision.

The Marketer's Path (about 6 hours)

For whoever will run the programme day to day. Ends with a 30-day starting plan.

1. Full GEO pillar page — 45 min. The vocabulary and framework everything after this assumes you already know.
2. GEO Glossary, skim key terms — 20 min. Front-loads the jargon in the external sources next.
3. Google AI features guidance and FAQPage deprecation notice [PRIMARY] — 20 min. Two documents, two common vendor pitches disarmed.
4. GEO paper in full, arXiv:2311.09735 [PEER-REVIEWED] — 45 min. Short, the actual origin of the term, and it contains the tactic that hurt visibility, which nobody selling GEO mentions.
5. Reuters Institute, Trends and Predictions 2026 [INDEPENDENT] — 30 min. Your source for any stakeholder conversation about scale.
6. Semrush AI Overviews study writeup [VENDOR] — 20 min. Shows what kind of query triggers an AI Overview, which changes what you build first.
7. Profound citation-source analysis writeup [VENDOR] — 20 min. The owned-versus-earned split reshapes a content calendar.
8. Otterly.ai's three studies — dead citations, deindexing, year-in-title [VENDOR, experimental] — 30 min. The field's best negative-result reporting; stops you repeating two expensive mistakes.
9. Full GEO Methodology page — 40 min. The operational detail behind the five phases already skimmed.
10. Tools section on this page — pick two to trial — 30 min. You cannot measure a channel with no instrument.
11. Templates section — build your first prompt set — 40 min. The one artifact every other measurement depends on.
12. GEO FAQ, skim — 20 min. Where you will point stakeholders asking what you just learned.

30-day starting plan:

• Week 1. Audit robots.txt and rendering. Build the prompt set template from this page. Baseline share of answer across the six surfaces by running the prompts manually and logging results.
• Week 2. Audit entity consistency: canonical name, sameAs links, and the one-sentence description across your site, LinkedIn, Crunchbase and any industry directories. Fix the worst three mismatches.
• Week 3. Pick five buyer prompts from the baseline where you were absent and map their likely fan-out sub-questions. Draft one answer-first page against the highest-value sub-question.
• Week 4. Re-run the prompt set. Compare against the Week 1 baseline. Decide, with the numbers in front of you, whether to expand the content programme or fix access issues first.

flowchart TD
    A["1-2. Pillar page and glossary - 65 min"] --> B["3-4. Google docs and GEO paper - 65 min"]
    B --> C["5-8. Reuters, Semrush, Profound, Otterly - 100 min"]
    C --> D["9. Methodology page - 40 min"]
    D --> E["10-11. Tools and templates - 70 min"]
    E --> F["12. FAQ skim - 20 min"]
    F --> G["30-day starting plan"]

The Marketer's Path: twelve steps across roughly six hours, ending in a concrete four-week plan rather than a reading list.

The Engineer's Path (about 4 hours)

For whoever owns robots.txt, rendering and schema. Ends with a technical checklist.

1. "Retrieval" and "How AI discovers your company," GEO pillar page — 20 min. Separates the training-agent decision from the retrieval-agent decision before you touch a config file.
2. OpenAI bots documentation [PRIMARY] — 20 min. GPTBot, OAI-SearchBot, ChatGPT-User and OAI-AdsBot do four jobs; conflating them is the most common self-inflicted visibility loss.
3. Anthropic crawler support article [PRIMARY] — 15 min. ClaudeBot, Claude-User and Claude-SearchBot need the same separation, plus non-standard Crawl-delay support.
4. Perplexity bots guide [PRIMARY] — 15 min. Perplexity-User deliberately ignores robots.txt on user-triggered fetches — easy to miss in your own logs.
5. Google AI features guidance plus Article, Organization, Product and Dataset docs [PRIMARY] — 30 min. The only schema types worth shipping in 2026; FAQPage is not one of them.
6. Gemini API grounding documentation [PRIMARY] — 20 min. url_citation offsets explain why Gemini quotes one sentence and not the next.
7. Anthropic web search tool documentation [PRIMARY] — 15 min. cited_text is a literal span lifted from your page — the case for self-contained sentences.
8. IndexNow specification [PRIMARY] — 15 min. A short integration that notifies Bing, and by extension Copilot, the moment a page changes.
9. Schema.org, Organization and Product type pages [PRIMARY] — 20 min. The vocabulary underlying every structured-data recommendation in this hub.
10. llms.txt specification [PRIMARY, unverified adoption] — 10 min. It is a proposal, not a standard any major vendor documents consuming — know that before shipping one.
11. Self-checks section on this page — run all against your own domain — 40 min. Converts documentation into an actual audit of your own site.
12. Crawler access decision matrix template — fill it in — 20 min. The artifact you hand to whoever owns the CDN or WAF configuration.

Technical checklist:

• robots.txt allows OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and bingbot, regardless of your position on training agents.
• Training-agent decisions (GPTBot, ClaudeBot, Google-Extended) are documented as a deliberate content-licensing choice, not an accidental blanket block.
• Key facts render in raw HTML, not only after client-side JavaScript execution.
• Organization structured data on the homepage includes a complete sameAs array.
• No FAQPage schema is being shipped on the assumption it produces a rich result.
• IndexNow is wired into the publishing pipeline.
• A CDN or WAF bot-mitigation rule is not silently returning 403s to documented AI crawler user agents.

flowchart TD
    A["1. Pillar retrieval sections - 20 min"] --> B["2-4. OpenAI, Anthropic, Perplexity bot docs - 50 min"]
    B --> C["5-7. Google schema and grounding docs, Claude search docs - 65 min"]
    C --> D["8-10. IndexNow, Schema.org, llms.txt - 45 min"]
    D --> E["11-12. Self-checks and decision matrix on your domain - 60 min"]
    E --> F["Technical checklist"]

The Engineer's Path: twelve steps across roughly four hours, ending in a checklist to run against your own robots.txt and schema before the marketing team asks why nothing changed.

Primary sources — official documentation

The vendors' own words about their own systems. Read these before any secondary summary of them, including ours.

Google Search Central — AI features guidance (developers.google.com/search/docs/appearance/ai-features). [PRIMARY] What it says: no additional technical requirements exist to appear in AI Overviews or AI Mode beyond ordinary Search eligibility — indexed, snippet-eligible, no special file or schema. Describes query fan-out and the practices that carry over from standard Search: crawlable robots.txt, internal linking, text-first content, structured data matching visible text, current Business Profile data. Why it matters: dismantles most of the "special AI optimization" pitch. The most-cited source in this hub, for good reason.

Google Search Central — FAQPage documentation (developers.google.com/search/docs/appearance/structured-data/faqpage). [PRIMARY] What it says: FAQ rich results stopped appearing in Google Search as of 7 May 2026, with rich result reporting and Rich Results Test support withdrawn through June 2026 and Search Console API support removed in August 2026. Why it matters: anyone recommending FAQPage schema as a 2026 visibility tactic has not read this page.

Google Search Central — Article, Organization/logo, Product and Dataset structured data documentation. [PRIMARY] What it says: these are the structured data types Google continues to document and support for entity clarity, business identity, product data and dataset discovery, each with defined properties. Why it matters: the actual schema worth shipping in 2026, in contrast to deprecated FAQPage.

Gemini API grounding documentation (ai.google.dev/gemini-api/docs/grounding). [PRIMARY] What it says: describes the flow by which the model decides a search is needed, generates and runs queries, synthesizes results, and returns inline url_citation annotations with character offsets into the generated text. Why it matters: the only documented explanation of how a Gemini sentence ties back to a specific source passage.

OpenAI bots documentation (developers.openai.com/api/docs/bots). [PRIMARY] What it says: defines GPTBot (training), OAI-SearchBot (surfaces sites in ChatGPT search), ChatGPT-User (user-triggered fetch) and OAI-AdsBot (ad landing validation, not training), each with distinct robots.txt behaviour. Why it matters: the training-versus-retrieval distinction, documented in OpenAI's own words.

OpenAI ChatGPT search help article (help.openai.com). [PRIMARY] What it says: ChatGPT search rewrites a query into targeted queries sent to search partners, names Bing and Shopify without claiming exclusivity, and states any website can choose to appear in ChatGPT search. Why it matters: the direct source correcting the "submission mechanism" myth.

Anthropic crawler support article (support.claude.com/en/articles/8896518). [PRIMARY] What it says: documents ClaudeBot (training), Claude-User (user-directed fetch) and Claude-SearchBot (search quality), including ClaudeBot's support for the non-standard Crawl-delay directive. Why it matters: without this article, most sites cannot tell which Anthropic agent they are actually blocking.

Anthropic web search tool documentation (platform.claude.com). [PRIMARY] What it says: Claude decides autonomously whether to search, can search multiple times per turn, and always returns citations as web_search_result_location objects with url, title and a literal cited_text span. Why it matters: cited_text is a literal extract, not a paraphrase — the strongest case in this hub for writing self-contained, quotable sentences.

Perplexity bots guide (docs.perplexity.ai/guides/bots). [PRIMARY] What it says: PerplexityBot surfaces and links sites in Perplexity's index and does not train models; Perplexity-User, for user-triggered fetches, generally ignores robots.txt because the fetch is user-initiated. Why it matters: the robots.txt asymmetry between the two agents changes how you read a spike in unexplained Perplexity traffic.

IndexNow (indexnow.org). [PRIMARY] What it says: a protocol for instantly notifying participating engines of new or changed URLs, rather than waiting for a scheduled recrawl. Participants: Microsoft Bing, Naver, Seznam.cz, Yandex, Yep. Why it matters: the most under-used quick technical win for Bing, and by extension Copilot's inherited index.

Schema.org. [PRIMARY] What it says: the shared vocabulary every structured-data type in this hub is built on — Organization, Product, Article, Dataset, and the now-deprecated-for-rich-results FAQPage. Why it matters: the ground truth when a vendor-specific document does not answer a schema question.

llms.txt specification (llmstxt.org). [PRIMARY] What it says: a September 2024 proposal by Jeremy Howard of Answer.AI for a markdown file giving language models a curated site summary, adopted by some documentation tooling. Why it matters, honestly: tagged [PRIMARY] because it is the spec's own site, but no major AI vendor documents its crawlers actually consuming it. Read it to know what you would ship. Do not buy it as a visibility service.

Research and studies

What has actually been measured, by whom, at what scale, and what it does not prove.

GEO: Generative Engine Optimization — arXiv:2311.09735, presented at KDD 2024. [PEER-REVIEWED] Authors: Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande. Measured the effect of nine content modifications on source visibility in generative engine responses using GEO-bench, a purpose-built benchmark of diverse queries paired with web sources, published November 2023. Headline finding: adding quotations from relevant or expert sources produced the largest single gain, up to roughly 40%; keyword stuffing had a negative effect. Does not prove current 2026 ranking behaviour — it predates AI Mode and the current model generation, and the authors themselves note tactic efficacy varies by domain.

Journalism, Media and Technology Trends and Predictions 2026 — Reuters Institute. [INDEPENDENT] Measured AI Overview prevalence, ChatGPT weekly active users, and organic referral changes to news publishers, using aggregated 2026 industry data including Chartbeat referral data across 2,500+ news sites, November 2024 to November 2025. Headline finding: AI Overviews appear in roughly 10% of US search results; referrals to news sites fell 33% globally and 38% in the US; Google still delivers roughly 500 times more referral traffic than ChatGPT from search alone. Does not prove causation for any single publisher, or that the trend is linear going forward — publisher expectations of further loss are surveyed opinion, not measured outcome.

Gartner search-volume forecast, press release, 19 February 2024. [INDEPENDENT, forecast not measurement] Measured nothing directly — it is an analyst forecast, published February 2024, predicting traditional search volume would fall 25% by 2026. Does not prove the actual 2026 outcome matched. It is cited across the industry as if it were a measurement; it is a prediction made two years before the date it predicts.

Semrush AI Overviews study, n=200,000 US keywords, collected September 2024. [VENDOR, disclosed methodology] Measured overlap between AI Overview citations and organic top-10 rankings, average link count per Overview, and the search-volume and intent profile of triggering keywords. Headline finding: more than half of desktop AI Overviews did not link the #1 organic result; overlap with the top 10 was roughly 20–26%; about 82% of triggering keywords had under 1,000 monthly searches; roughly 80% were informational intent. Does not prove causation between any page attribute and citation — it is descriptive of one snapshot in one market.

Profound citation-source analysis, 11.84 billion citations, 3.02 million domains, 29 industries, April to July 2026. [VENDOR, large sample, unaudited] Measured the proportion of AI citations pointing to brand-owned domains, by engine. Headline finding: roughly 57% of AI citations globally point to brand-owned domains, ranging from about 47% on ChatGPT to 69% on Gemini. Does not prove independent verifiability — the figures come from a single vendor's own pipeline with no published audit. Large is not the same as verified.

Otterly.ai dead-citations study, over 20 million cited URLs, one-month snapshot, 2026. [VENDOR, descriptive not controlled] Measured the proportion of AI-cited URLs, across seven engines, returning dead or unreachable responses. Headline finding: 19.3% were dead; highest on ChatGPT at 25.1%, lowest on Google AI Overviews at 12.6%. Does not prove why any individual URL died or whether the rate holds over time — it is a snapshot, not longitudinal.

Otterly.ai AI-content deindexing experiment, two fresh domains, roughly 1,000 AI-generated posts each, 2026. [VENDOR, genuinely experimental, small n] Measured whether scaled AI-generated publishing triggers algorithmic deindexing. Headline finding: both domains were deindexed with no manual action notice; one site's daily impressions fell from 1,629 to 15. Does not prove a general law for any scale, domain age or niche — two sites supports "this can happen," not "this will happen to you."

Otterly.ai "year in title" test, 11 pages, two waves, 2026. [VENDOR, negative result] Measured whether adding the current year to page titles increases citations. Headline finding: an apparent 56–61% growth — until a single outlier page turned out to drive 86–93% of the gain, and an untouched control with the year already in its title rose just as much. The authors concluded the effect could not be separated from noise. This is the most valuable study on this page precisely because it is a negative result honestly reported: a vendor admitting "we could not prove it worked" is rare, and it is a strong signal about that outfit's credibility on everything else it publishes.

Tools

Ahrefs Brand Radar measures brand and competitor mentions across the widest surface set here, including social platforms, but not why a mention happened, and its Grok/social coverage is newer than its rank-tracking core. Suits teams already running Ahrefs who want one added module.

Otterly.ai is the most affordable entry point and the source of the field's best negative-result publishing, which says something about its own credibility too. Measures presence and citation across six surfaces cheaply, without Evertune's deep sampling. Suits anyone starting from zero budget.

Profound covers the widest engine list, including DeepSeek and Grok, and produced the largest citation dataset in this hub. Sales-gated with no public pricing. Suits mid-market and enterprise teams who need breadth and can absorb an unpublished price.

Peec AI is narrower — three engines — but the only tool here that separates "used as a source" from "cited with a link." Does not cover AI Overviews or Copilot. Suits teams whose buyers concentrate on ChatGPT, Perplexity and Gemini.

Evertune's edge is methodological: sampling each prompt 100 times across 11 models addresses the non-determinism that undermines single-run measurement everywhere else. Sales-gated, enterprise-priced. Suits teams that prioritize statistical rigor over feature breadth.

Scrunch adds AI bot crawl-behaviour monitoring and persona/geography benchmarking on top of citation tracking. The 7-day trial is short for a channel with weeks-to-months feedback cycles. Suits teams wanting server logs correlated with citation outcomes.

Rankscale claims the widest engine count, 17+, and an "AI readiness score" from 200+ factors, but the scoring methodology is not publicly documented and credit-based pricing makes cost hard to forecast. Suits teams trading transparency for breadth, knowingly.

Semrush AI Visibility is lowest-friction for teams already paying for Semrush, bundling prompt-level visibility into an existing subscription, with narrower coverage than dedicated platforms above. Suits convenience over specialization.

No independent third-party accuracy audit of any tool here exists that we could verify. Every number comes from the vendor's own materials. We do not resell any of these tools, and every price above will be out of date by the time you read this.

Free diagnostic checks anyone can run today

No subscription required for any of these. Run them before you buy anything.

1. Fetch and read your robots.txt. curl -s https://yourdomain.com/robots.txt Look for Disallow against OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot or bingbot — a self-inflicted retrieval block, separate from whatever you decide about GPTBot, ClaudeBot or Google-Extended.

2. Impersonate each documented AI user agent. curl -A "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.4; +https://openai.com/gptbot" -I https://yourdomain.com/ Repeat with the OAI-SearchBot and PerplexityBot strings on the GEO pillar page's crawler table. A 403 from your CDN or WAF, not robots.txt, is a silent block most teams never find.

3. Compare rendered HTML to raw HTML. curl -s https://yourdomain.com/your-page | grep -i "<title>" Compare against a browser view-source after JavaScript executes. If key facts only appear after client-side rendering, a text-only crawler may never see them.

4. Run the Rich Results Test. Visit search.google.com/test/rich-results on your homepage and one product or article page. Confirm Organization and Article/Product markup validate. Do not submit FAQPage expecting a rich result.

5. Check Bing Webmaster Tools indexation. Log in and check the Site Explorer indexation report. Given Bing's role as a ChatGPT search partner and Copilot's inherited index, this is a two-engine check disguised as one.

6. Run a fixed prompt set manually across all six surfaces and log it. Build a spreadsheet named GEO-prompt-tracker-[yourcompany].xlsx with columns: prompt, engine, date run, brand mentioned (yes/no), cited with link (yes/no), competitors mentioned, response excerpt. Run 10–15 prompts by hand in Google AI Mode, ChatGPT, Gemini, Claude, Perplexity and Copilot — the manual version of what every paid tool automates.

7. Check your Knowledge Panel. Search your own company name in Google. Note whether the panel's logo, description and sameAs links match your site, or whether a competitor's panel appears due to entity confusion.

8. Audit sameAs consistency. Open every profile URL in your Organization schema's sameAs array. Confirm name, logo and one-line description match your site word for word.

9. Check for dead citations. Search your brand name in each AI surface, collect the linked citations, and check each URL's status: curl -o /dev/null -s -w "%{http_code}\n" https://example.com/cited-page Given the 19.3% cross-engine dead-citation rate Otterly.ai observed, expect to find at least one.

10. Verify IndexNow is wired up. curl -s "https://www.bing.com/indexnow?url=https://yourdomain.com/your-page&key=yourkey" A successful response confirms the integration; a missing key means it is not set up yet.

Templates and frameworks

Copy any of these directly. They are deliberately plain markdown so they drop into a spreadsheet or a wiki without reformatting.

Prompt-set template. Columns: prompt text · prompt type · target engine · date run · brand mentioned · cited with link · competitors named · response excerpt · notes. Prompt-type taxonomy: informational ("what is GEO"), comparison ("X vs Y"), listicle-style ("best agencies for..."), transactional ("who should I hire for..."), branded ("is [company] good at..."). Mix all five types; a prompt set built only from branded prompts overstates your visibility.

Share-of-answer tracking sheet. Columns: date · engine · prompts run · prompts mentioning brand · share of answer % · prompts citing with link · citation rate % · top competitor mentioned · competitor share %. Recompute monthly from the same prompt set, sampling each prompt more than once per run given documented response variance.

Competitor benchmark grid. Rows: your brand plus up to five named competitors. Columns: one per engine (AI Overviews, ChatGPT, Gemini, Claude, Perplexity, Copilot), cell value = share of answer % on the shared prompt set. Add a final column for citation-tier mix (owned/verified/earned) per competitor where observable.

Canonical entity description worksheet. Rows: every property that describes your company (site, LinkedIn, Crunchbase, industry directories, press bios, Wikidata if applicable). Columns: current description · word-for-word match to canonical (yes/no) · date last checked · owner · fix logged. The goal is a single description repeated near-verbatim everywhere, not five slightly different ones.

Crawler access decision matrix. Rows: every documented agent (GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot, ClaudeBot, Claude-User, Claude-SearchBot, PerplexityBot, Perplexity-User, Googlebot, Google-Extended, bingbot). Columns: training or retrieval · current robots.txt status · desired status · business reason for the decision · owner · last verified. This is the single artifact that prevents the most common unforced error in this field.

Monthly AI visibility report outline. Sections: executive summary (one paragraph, plain language) · share of answer by engine, with month-over-month change · citation rate and citation health (percentage of live links) · competitive share on the shared prompt set · notable framing or sentiment shifts · entity and structured-data changes shipped this month · access or technical issues found and fixed · next month's priorities. Keep it to two pages. A report nobody reads protects nobody's visibility.

What to be skeptical of

A checklist of claims that should slow you down, with the counter-evidence.

• "We guarantee AI rankings or citations." No one can. Generated answers are non-deterministic, ranking systems for every engine except the disclosed parts of Google's process are undocumented, and model versions change without notice. The GEO pillar page's FAQ makes the same point about Artlogic's own work.

• "llms.txt is a paid service we'll implement for you." llms.txt is a free, one-hour-to-ship markdown file specified at llmstxt.org. No major AI vendor documents consuming it. Paying a premium for it as a visibility service is paying for a hypothesis with no vendor confirmation behind it.

• "FAQ schema is a 2026 AI visibility tactic." Google's own documentation states FAQ rich results stopped appearing in Google Search on 7 May 2026, with support being withdrawn through August 2026. This is a 2023 tactic being resold as current.

• "Submit your site to ChatGPT." There is no submission mechanism. OpenAI's own help documentation states any website can appear in ChatGPT search organically; what you actually control is OAI-SearchBot access and Bing indexation.

• "We tested it once and it worked." A single prompt run on a single day is an anecdote, not a measurement, given documented response variance across sessions and model versions. Evertune's 100-samples-per-prompt approach exists specifically to address this. Otterly.ai's own "year in title" test looked like a 56–61% win until a second look showed it was noise.

• "You can pay to appear in organic AI citations." Organic citations are not a purchasable inventory. Labelled advertising products inside AI surfaces exist and are disclosed as advertising. If a vendor claims otherwise, ask them to name and document the mechanism.

• "Our tool's accuracy is independently audited." No independent third-party accuracy audit of any AI-visibility tool exists that we could verify, including the eight tools reviewed on this page. Treat every number from every one of them, ours included, as vendor-reported until proven otherwise.

Continue in the Knowledge Hub

• GEO Overview — the pillar page: what GEO is, how AI actually discovers a company, and the six surfaces compared side by side.
• GEO Methodology — the five phases in operational detail, including where measurement fits.
• GEO Glossary — sixty-nine defined terms, including share of answer and prompt set, used throughout this page.
• GEO FAQ — sixty questions, including several about the tools and studies referenced here.
• GEO Case Studies — how these sources and tools get applied inside real engagements.

Related services: AI Visibility & GEO · Search Dominance

Work with Artlogic

Every source, study and tool in this library is here because we use it, tested it, or found it worth reading against its own incentive. If you would rather have someone run the audit, build the prompt set, and read the vendor data for you, that is what Artlogic's AI Visibility Audit does first.

Book a strategy call

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "CollectionPage",
      "@id": "https://artlogic.ca/geo/resources#page",
      "name": "GEO Resource Library",
      "description": "A curated GEO knowledge library: primary vendor documentation, peer-reviewed research, honest tool reviews, self-diagnostic checks, and three role-based reading paths.",
      "url": "https://artlogic.ca/geo/resources",
      "datePublished": "2026-08-06",
      "dateModified": "2026-08-06",
      "inLanguage": "en",
      "isPartOf": { "@id": "https://artlogic.ca/geo#hub" },
      "author": { "@id": "https://artlogic.ca/#organization" },
      "publisher": { "@id": "https://artlogic.ca/#organization" },
      "mainEntity": {
        "@type": "ItemList",
        "name": "GEO primary sources, research and tools",
        "itemListElement": [
          { "@type": "ListItem", "position": 1, "name": "Google Search Central — AI features guidance", "url": "https://developers.google.com/search/docs/appearance/ai-features" },
          { "@type": "ListItem", "position": 2, "name": "Google Search Central — FAQPage documentation", "url": "https://developers.google.com/search/docs/appearance/structured-data/faqpage" },
          { "@type": "ListItem", "position": 3, "name": "Gemini API grounding documentation", "url": "https://ai.google.dev/gemini-api/docs/grounding" },
          { "@type": "ListItem", "position": 4, "name": "OpenAI bots documentation", "url": "https://developers.openai.com/api/docs/bots" },
          { "@type": "ListItem", "position": 5, "name": "OpenAI ChatGPT search help article", "url": "https://help.openai.com" },
          { "@type": "ListItem", "position": 6, "name": "Anthropic crawler support article", "url": "https://support.claude.com/en/articles/8896518" },
          { "@type": "ListItem", "position": 7, "name": "Anthropic web search tool documentation", "url": "https://platform.claude.com" },
          { "@type": "ListItem", "position": 8, "name": "Perplexity bots guide", "url": "https://docs.perplexity.ai/guides/bots" },
          { "@type": "ListItem", "position": 9, "name": "IndexNow", "url": "https://www.indexnow.org" },
          { "@type": "ListItem", "position": 10, "name": "Schema.org", "url": "https://schema.org" },
          { "@type": "ListItem", "position": 11, "name": "llms.txt specification", "url": "https://llmstxt.org" },
          { "@type": "ListItem", "position": 12, "name": "GEO: Generative Engine Optimization (arXiv:2311.09735)", "url": "https://arxiv.org/abs/2311.09735" }
        ]
      }
    },
    {
      "@type": "Organization",
      "@id": "https://artlogic.ca/#organization",
      "name": "Artlogic",
      "url": "https://artlogic.ca"
    },
    {
      "@type": "BreadcrumbList",
      "itemListElement": [
        { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://artlogic.ca" },
        { "@type": "ListItem", "position": 2, "name": "GEO", "item": "https://artlogic.ca/geo" },
        { "@type": "ListItem", "position": 3, "name": "Resources", "item": "https://artlogic.ca/geo/resources" }
      ]
    }
  ]
}

Schema note. CollectionPage with a mainEntity ItemList is used here rather than FAQPage, even though this page contains no FAQ content, to make explicit that this is a curated collection rather than a set of question-answer pairs Google would consider for FAQ rich results, a format that stopped appearing on 7 May 2026 in any case.

Last reviewed 6 August 2026. Every credibility tag reflects our own judgment of the source's incentive, not an external certification. Pricing figures for tools will drift; check the vendor's current page before budgeting.