Frequently asked questions

Making a site something an AI system can actually read and use — not just something that ranks in a search engine. A crawler like GPTBot or PerplexityBot needs to be able to reach your pages, your content needs to render without JavaScript, and your structured data needs to tell a machine unambiguously who you are. GEO is the AI-facing counterpart to SEO: the goal shifts from ranking higher in a results page to being reachable and legible to the systems that increasingly stand in for search.

SEO optimizes for a search engine's ranking algorithm and a human clicking a link in a results page. GEO optimizes for a different reader entirely: an LLM crawler indexing content, an answer engine like ChatGPT or Perplexity fetching a page live to compose an answer, or a coding agent pulling documentation. The two overlap — fast, well-structured pages help both — but diverge in places SEO never had to consider: whether a page needs a browser to render its real content at all, since most AI crawlers never wait for JavaScript, or whether robots.txt blocks a bot that didn't exist when the file was last edited.

The two terms are used almost interchangeably in practice. Where people do draw a distinction, AEO tends to emphasize getting cited inside a generated answer, while GEO covers the broader question of whether an AI system can read a site at all. OSEM focuses on the accessibility side of that spectrum — whether your site is technically reachable and legible to an AI agent — rather than the separate, much harder to control question of whether a given answer chooses to cite you.

Because a growing share of how people find and evaluate a business now runs through an AI system instead of a search results page — a chatbot answering a question, a shopping agent comparing options, a coding agent reading documentation on someone's behalf. If GPTBot, ClaudeBot, or PerplexityBot can't fetch your pages, or your real content only appears after JavaScript finishes running, that whole channel doesn't see a lower-ranked page — it sees nothing. A site can look complete to a human visitor and be functionally invisible to the systems now doing a meaningful share of that visitor's research.

llms.txt is a plain-text file at your site's root — a short, plain-language description of what the site is, what it offers, and links to its key pages — written for an LLM to read directly, the same way robots.txt is written for crawlers. It's optional and still uncommon; most sites, including most competitors in this space, don't have one, which makes it a low-cost, high-signal addition. It gives an AI agent a direct, unambiguous summary instead of making it infer one from marketing copy. OSEM's Fix Kit generates one tailored to the site being scanned.

The named ones this site's own robots.txt already allows: GPTBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended, CCBot, Bytespider. That's the same list a scanned site is checked against for crawler access, and the same list checked live during the render-gap check.

Yes, and it's one of the more common findings once you actually check. Google's classic search index and its AI features are governed by different signals than the crawlers this product checks — Google-Extended, for instance, is separate from regular Googlebot indexing — so a page can be fully indexed for search while still failing basic AI-agent accessibility: blocked for a specific bot in robots.txt, or rendering its real content only after JavaScript runs. Ranking well and being AI-reachable are correlated, not identical.

Five things, run against your homepage: crawler access (does robots.txt allow known AI crawlers), content clarity (is there a real <h1> and real body text in the raw HTML, before JavaScript runs), entity signal (structured data, a clear site name, an authorship signal, a canonical link), render gap (how much content only appears once a real browser renders the page), and page speed (how fast the page responds). It returns one overall score, plus a note on how many other pages it found linked from your homepage — not a per-check breakdown, which is reserved for the full scan.

The free scan only ever looks at your homepage. The full scan crawls your actual site — same-domain links, breadth-first from the homepage — and builds a site map: every discovered page, its own score, and an expandable per-check breakdown, instead of one number for the whole site. It also unlocks the per-check findings and, once you own the Fix Kit, the AI citation check.

The free homepage scan finishes in under 60 seconds — it's a small, fixed set of checks against one page. The full site-wide scan runs as a background job, since it has real work to do: crawling up to 100 pages and rendering the first 5 of them in a real headless browser for the render-gap check. You get a job ID immediately and the page polls it until the crawl finishes, rather than holding the request open the whole time.

Up to 100 pages, 3 clicks deep from the homepage by default, same-domain links only. The render-gap check — the expensive one, a real headless-browser page load each — only runs on the first 5 pages scanned; the other checks run on every page the crawl is allowed to fetch.

Yes. It's fetched once per site and reused for every page in the crawl. A page robots.txt disallows still shows up in the site map, so you can see that it's unreachable, but it's never actually fetched — and it never gets a page-speed score either, since that check requires an actual request.

Not for the free homepage scan — no card, no signup, one per IP per day. The full site-wide scan is gated behind an account, since it runs as a real background crawl job rather than a single request, and its results (site map, citation check) are tied to that account going forward.

A single number is what nearly every free scanner already gives away. A score tells you something's wrong; a map tells you exactly which of your pages an AI agent can and can't reach — the thing you'd actually need in hand to go fix anything. That's the deliverable OSEM leads with, not the number.

Whether robots.txt lets GPTBot, ClaudeBot, PerplexityBot, and the other named AI crawlers reach your pages at all. Most existing robots.txt files were written before these crawlers existed, and block them by accident.

What's actually in the raw HTML an agent receives — a real <h1>, real body text — versus content that only appears once JavaScript runs, which most crawlers never wait for.

Whether a page needs a browser to look complete. We render it the way an agent would and compare that against the raw HTML it gets on the first request.

Whether there's an unambiguous answer to “who is this” anywhere a machine can parse it — Organization schema, a clear about section, one consistent brand name — so an agent doesn't have to guess.

How fast your site responds to a crawler at all. A slow origin gets less crawl budget, so a bounded crawl like ours reaches fewer of your pages in the same time.

A weighted average of the five checks above, normalized to a 0–100 scale. Render gap, entity signal, and page speed each count for more than crawler access and content clarity — the checks that tend to point at a harder, structural problem carry more weight than the ones that are comparatively easy to patch.

The product's own status coloring draws the line at 80 and 50: 80 and above reads as a pass, 50–79 as a warning worth addressing, and below 50 as a real problem an AI crawler is likely to actually hit. Those aren't arbitrary marketing thresholds — they're the same cutoffs the app itself uses to color a score green, amber, or red.

Yes, on the full site-wide scan. A site that's already fully accessible can also earn a citation bonus — a small number of points added for every citation-tracking query where an AI answer actually cited it, capped at a modest total — on top of the 0–100 accessibility score. It's deliberately additive-only, never a penalty for not being cited, since that's driven by off-site signals outside this product's control, and deliberately uncapped past 100, so a site that's both perfectly accessible and already getting cited visibly shows it instead of the bonus being discarded at a ceiling.

A corrected robots.txt, an llms.txt generated specifically for your site, and JSON-LD/meta-tag snippets closing the entity-signal gap the scan found — plus a full technical report ranking every issue by what it actually costs you, and a step-by-step install guide tailored to your platform. It also unlocks the AI citation check and a score-gap report. $500, one-time.

No. The Fix Kit generates the files and instructions; you or your developer installs them. OSEM doesn't get write access to your site or push changes to it directly — that's a deliberate boundary, not a current limitation.

A side-by-side of two numbers: the score you'd reach by installing the Fix Kit's generated files yourself, versus the score reachable with full Managed Remediation. It's what makes the upsell to Managed Remediation concrete instead of a generic pitch — you can see exactly how many points are locked behind judgment calls a generated file can't make for you.

Shopify, WordPress, Websupport, or a generic/custom setup — the guide is written for whichever one the scanned site is actually built on.

It covers the fixes that need human judgment rather than a generated snippet — content structure, information architecture, entity clarity — reviewed and implemented by a person against your specific scan and site map. It's a custom-scoped, quoted engagement rather than a fixed SKU, and it includes a free weekly re-scan by email while the engagement is being scoped.

Yes — the AI citation check (gated behind the Fix Kit purchase, since each check costs a real API call) and the score-gap report both come with it, on top of the files themselves.

Three tiers. The scan itself is free — one homepage scan per IP per day, no card, no signup. The Fix Kit is a $500 one-time purchase that unlocks the full site-wide crawl, the AI citation check, and the generated fix files. Managed Remediation is a custom quote, scoped once a human has reviewed your scan and map. See /pricing for the full feature-by-feature comparison.

No — it's priced as a one-time $500 purchase specifically because it delivers a fixed set of files and a report, not an ongoing service. Billing it monthly would only teach a customer to grab the files and cancel after the first month, so the genuinely ongoing part — monitoring — is sold separately as its own optional add-on instead of bundled into a subscription no one would keep.

Periodic re-scans and refreshed citation checks, plotted as a score-over-time trend — replacing the one-off results page with a persistent view of whether your fixes are holding and whether you're showing up in AI answers over time.

No — monitoring is an add-on within the Fix Kit tier, since it re-scans and re-checks the same things the Fix Kit purchase unlocks (the full crawl and the citation check) rather than standing alone as a separate product.

Yes — monitoring is billed monthly and can be canceled independently at any time. The one-time Fix Kit purchase itself (the generated files and report) stays yours either way, since only the ongoing re-scan/monitoring part is what you'd be canceling.

SMB and mid-market businesses, roughly $1M–$50M in revenue, where an AI agent reading the site directly has real operational stakes — DTC e-commerce sites an agentic shopping tool might browse, B2B SaaS companies whose docs a coding agent reads, professional-services firms, and content publishers. These are businesses with enough site complexity that a full crawl finds real problems, and enough at stake to act on what it finds.

The scan itself works on any site, on any platform — it only ever reads the actual HTTP response and rendered page a crawler or browser would get, regardless of what generated it. The Fix Kit's install guide is written specifically for Shopify, WordPress, Websupport, or a generic/custom setup, so however your site is built, the instructions match it.

No. The files themselves (robots.txt, JSON-LD, llms.txt) are meant for a developer or platform admin to install, but the Fix Kit ships with a step-by-step guide written for your specific platform rather than assuming you can hand-edit server config. If installing them yourself still isn't realistic, that's exactly the gap Managed Remediation is there to close.

A famous brand's AI-answer visibility is often driven by ambient signals — press coverage, industry news, a long public track record — that exist independently of how technically crawlable its own website is, so a large recognizable brand can score poorly on-site and still get named in an AI answer anyway. That decouples the fix from the outcome in a way that makes for a slow, low-urgency sales conversation. SMB and mid-market businesses without that ambient reputation don't have the same fallback, so fixing on-site accessibility has an outsized, directly provable effect on whether they're usable by AI at all — which is where OSEM focuses today.

No, and it's worth being direct about why. OSEM can guarantee your site is technically accessible to AI crawlers and agents — that's fully within the product's control and what remediation actually fixes. Whether a specific AI answer engine chooses to cite you in a given answer depends heavily on off-site signals — backlinks, Reddit mentions, Wikipedia presence — that no on-site fix controls. A citation-tracking result can stay flat even after a full remediation, and that's expected, not a failure of the fix.

Because citation is driven mostly by third-party authority signals rather than a site's own technical quality. OSEM stays scoped to what's actually provable and fixable: on-site accessibility. Off-site authority-building is a genuinely different skillset, closer to PR or content marketing than technical remediation, and it isn't bundled into what remediation promises today.

A branded query names your business directly ("is [business] any good"); a discovery query only names your category and location, never your business ("best plumber in [city]"). Getting cited on a branded query just confirms the AI engine has heard of you. Getting cited on a discovery query means it recommended you unprompted — the real test of whether you show up when a prospective customer who's never heard of you asks.

A name mention alone isn't treated as a real hit. The check only counts it once the AI's answer also includes a corroborating detail pulled from your own business profile — your category, part of your location, or a founder's name — so a same-named but unrelated business doesn't register as a match. If there's nothing in your profile to corroborate against, an ambiguous mention is treated as not confirmed rather than assumed accurate.

Yes. Queries are generated in the language your site actually declares — read from its HTML lang attribute — falling back to a lookup by domain (its country-code TLD), and to English only if neither is available. A Slovak plumber's prospective customers ask ChatGPT in Slovak, not English, so that's the language the check actually queries in, not a translated English question.

The URL you submit and the pages the crawler discovers on that domain; a best-guess business profile detected from the scanned site, used to generate the citation-check queries; your account email, for login and opt-in results; and your scan history, if you're signed in. None of it is sold, and it isn't shared beyond what's needed to run the scan itself. Full detail lives on the Privacy & Cookie Policy page.

There's no fixed universal interval, but two built-in cadences give a sense of scale: the optional monitoring add-on re-scans and re-checks citations on an ongoing basis to catch drift over time, and a lead in the middle of scoping a remediation engagement can get a free weekly re-scan by email (opt-in only, never a default) to see progress before committing to anything. Outside of a formal plan, re-running a scan after any real content, platform, or hosting change is the simplest rule of thumb.