Terminology
What do AEO, GEO, AI Visibility, and LLM SEO actually mean?
They’re overlapping labels for one goal: getting your brand surfaced, cited, and recommended inside AI answers rather than only in traditional search links.
- AI Visibility — the umbrella outcome: how often and how prominently your brand appears across AI answers.
- AEO (Answer Engine Optimization) — optimizing to be the cited answer in AI-powered answer features like Google AI Overviews and featured snippets.
- GEO (Generative Engine Optimization) — optimizing to be retrieved and cited by generative assistants like ChatGPT, Claude, Perplexity, and Gemini. The term was coined in a Princeton/Georgia Tech/Allen Institute paper at ACM SIGKDD 2024.
- LLM SEO / LLMO / AIO — loose synonyms for the same discipline.
The line between AEO and GEO is blurry and the two are often used interchangeably; the industry hasn’t standardized on one term, with “AI visibility” emerging as the catch-all (Jasper, Wikipedia: GEO).
Why it matters
How big is AI search now?
ChatGPT reached roughly 800–900M weekly active users in early 2026, up from ~400M a year earlier (Omnibound). Directional query-share estimates put Google around 80% of queries, ChatGPT ~17%, and other AI/alt-search ~3% — though these splits come from vendor roundups, not a single audited source, so treat these as directional trendlines (QuickSEO).
What’s happening with “zero-click” search?
Zero-click rates hit record highs in 2025 — roughly 58–60% of US/EU searches ended without a click to an external site, and reportedly ~83% when an AI Overview was present. AI Overviews are reported to appear in ~25% of Google searches (up from ~13% in early 2025), with some publishers reporting 20–40% organic-traffic declines (Omnibound, Superlines). These are vendor-aggregated, so please confirm the underlying Pew/Bain/SparkToro figures before putting hard numbers in front of a client. The core point holds regardless of the exact percentage: more and more answers happen without a visit to your site.
Does AI traffic actually convert?
The recurring theme is “less volume, higher quality.” Holiday 2025 reporting showed AI referral traffic to US retail growing several-hundred percent year-over-year and converting meaningfully better than non-AI traffic (Omnibound, Goodie). Exact figures vary by source and methodology.
How AI engines choose what to cite
How do AI answers decide which sources to use?
Modern AI search is increasingly “agentic RAG”: a single question is broken into many sub-queries (“query fan-out”), passages are retrieved and scored for each, and the best are synthesized into one answer. Google’s AI Mode uses a custom model built for fan-out (iPullRank on fan-out, Search Engine Land). Most retrieved pages are never cited, so being retrievable isn’t enough — you have to be the passage worth quoting.
What content actually drives citations?
The peer-reviewed Princeton GEO study (10,000 queries) found the strongest levers were adding statistics, citing sources, and adding quotations, plus authoritative, fluent writing — boosting visibility in generative answers by up to ~40%. Keyword stuffing performed actively worse than doing nothing.
Do brand mentions matter more than backlinks?
Evidence says yes. Ahrefs studied ~75,000 brands and found unlinked brand mentions correlated most strongly with AI visibility (~0.66), while raw backlinks correlated lowest (~0.22) — roughly a 3× gap. A follow-up found YouTube mentions correlated even higher. These are correlations, not proof of causation, but the signal is consistent: earned mentions across third-party sources (Reddit, Wikipedia, YouTube, review sites, press) matter more than link-building alone.
The llms.txt question
Should we publish an llms.txt file?
The honest 2026 answer: probably not for AI-search visibility. llms.txt is a markdown index file proposed in 2024 to help AI agents navigate a site. The “it boosts AI visibility” framing was added later by the SEO industry on speculation, and the evidence is skeptical:
- Ahrefs (June 2026, 137K domains): of domains publishing llms.txt, 97% of those files got zero requests in a month — and the bots that did fetch them were mostly coding/agent tools, not AI-search crawlers.
- SE Ranking (300K domains): found no link between llms.txt and citation frequency.
- Google doesn’t use it. John Mueller compared it to the discredited keywords meta tag, and Google’s own AI-optimization guide says special files like llms.txt aren’t needed.
The narrow exception is futureproofing for AI agents completing tasks on your site (navigation, transactions) — not visibility. Bottom line: low cost to publish, but don’t expect AI-search lift, and don’t let it crowd out the tactics that are evidence-backed.
Best practices
What’s actually worth doing in 2026?
Evidence-backed, in rough priority:
- Increase factual density — statistics, named citations, direct quotes, authoritative voice (Princeton GEO).
- Earn brand mentions across Reddit, Wikipedia, YouTube, review sites, and press (Ahrefs).
- Structure content for extraction — clear, self-contained sections that answer one question well.
- Clean structured data (schema) — Organization, Product, Article, FAQ; helps AI verify and attribute claims (Digidop).
- Keep facts fresh — current pricing, dates, and details; AI favors recent, specific, verifiable sources.
Avoid: keyword stuffing and promotional fluff (both correlate negatively with citations). Treat single-vendor “do X for Y% lift” claims skeptically unless independently replicated (evidence review).
Measurement
Why is AI visibility so hard to measure?
Because AI output is non-deterministic. SparkToro’s January 2026 study (Rand Fishkin, ~2,961 runs) found there’s less than a 1-in-100 chance that two responses to the same prompt return the same list of brands, and roughly 1-in-1,000 that they match in the same order. Tracking your “rank” in ChatGPT is, in Fishkin’s words, largely “baloney.”
So what can you measure?
Appearance rate / share of voice across many prompts and repeated runs. Even a skeptic like Fishkin concluded that while exact rankings are noise, how often a brand shows up across dozens-to-hundreds of runs is a statistically meaningful “consideration-set” signal. The right metric for clients is “we appear in X% of relevant AI answers,” not “we’re #3 in ChatGPT.” This is why credible tools (Grail included) run many prompts repeatedly and report rates and trends.
What else makes measurement messy?
- Prompt diversity — real users phrase questions very differently, so any fixed prompt set under-samples reality.
- No AI Overviews API — every vendor scrapes/simulates Google AI Overviews, and even the best detection is only ~68% reliable (Feb 2026 benchmark). All AIO data in this category is partial.
- Personalization — answers vary by account history, location, and device.
- Attribution gaps — AI answers often drive zero clicks, so influence happens with no analytics trail. Enterprise leaders tend to over-state their confidence here (Branch survey of 300 leaders).
The tool landscape
What tools exist, and what do they cost?
The market splits into mention/sentiment trackers (most of the field) and a small set that also verifies factual accuracy. Pricing as of June 2026 (re-verify before quoting — it changes often):
| Tool | Engines tracked | Notable for | Entry price |
|---|---|---|---|
| Profound | Up to 10 (tiered) | Enterprise leader; real-user “prompt volumes” panel | $99/mo (ChatGPT only) → $399 → custom |
| Peec AI | 3 base + add-ons | Europe/GDPR focus, agency workspaces | ~€205 → €675/mo |
| Otterly.AI | 4 base + add-ons | Cheapest with AI Overviews on every tier | $29 → $489/mo |
| Scrunch AI | ~8, all tiers | Crawler analytics; markets hallucination monitoring | $250 → $500/mo (acq. by Sitecore) |
| Goodie AI | Up to 11 | End-to-end AEO + revenue attribution | $399/mo → custom |
| Writesonic | Up to 10 (Ent.) | GEO + content generation + auto-fix “Action Center” | $79 → $399/mo |
| Trakkr | All 8 on every paid plan | Best low-cost coverage; genuine free tier | Free → ~$100–500/mo |
| Ahrefs Brand Radar | 6 (no Claude) | Ties AI visibility to Ahrefs’ link/search data | ~$828–1,148/mo (incl. base sub) |
| Semrush AI Toolkit | 5 (no Claude) | Strong brand sentiment + recommendations | $99/user/mo → bundles |
| SE Ranking | Google AIO/AI Mode + ChatGPT | Shows exact source URLs Google pulled into AIO | ~$52 → $489/mo |
| HubSpot AI Search Grader | 3 (no AIO) | Free one-time diagnostic | Free ($50/mo for monitoring) |
| Bluefish AI | 6 incl. Amazon Rufus | Factual-accuracy verification leader (enterprise) | Quote-only |
Sources: official pricing pages and 2025–26 third-party reviews including Trakkr reviews, Otterly, Semrush KB, HubSpot AEO Grader, Bluefish. Entry tiers are deliberately thin — typically 15–50 prompts and 1–3 engines; “all engines” usually means Enterprise.
Which tools cover Google AI Overviews?
Most self-serve tools do (Profound, Peec, Otterly, Scrunch, Goodie, Writesonic, Trakkr, Ahrefs, Semrush, SE Ranking) — but all via scraping/browser simulation, never an official API, because Google doesn’t offer one. HubSpot’s free grader does not cover live AI Overviews. Expect any AIO data, from any vendor, to be partial.
Which tools actually check whether AI is accurate about a brand?
This is the market’s clearest gap. Bluefish AI leads with a dedicated claim-by-claim verification module (enterprise, quote-only). Profound offers FactCheck as of July 2026. Knowatoa does a lighter “truthfulness” check, and Scrunch markets hallucination monitoring (depth unconfirmed). Everyone else — Peec, Otterly, Goodie, Writesonic, Trakkr, HubSpot, Ahrefs, Semrush, SE Ranking — tracks mentions, sentiment, and share-of-voice but not factual correctness. For a brand whose worry is “is AI saying the wrong thing about us,” accuracy verification is rare and largely absent below the enterprise tier — which is precisely where Grail Analytics focuses.
If you’re interested in learning more, sign up for a free seven day trial, or drop us a line if you have questions: hello@grailanalytics.ai.
Sources are linked inline. The highest-confidence, primary research underpinning Part 2: the SparkToro non-determinism study, the Ahrefs llms.txt and brand-mentions studies, the Princeton SIGKDD GEO paper, and Google’s own Mueller statements and AI-optimization guide. Traffic and market-share percentages are vendor-aggregated and directional — confirm primary sources before client use.