9 free checks to audit your AI visibility by hand

You do not need a paid platform to spot the biggest AI visibility problems. You do need a disciplined manual audit that checks whether AI systems can access your site, understand your brand, verify your claims, and repeat your facts accurately. That is the real starting point for Generative Engine Optimization (GEO).

The timing matters. SOCi's 2026 Local Visibility Index, based on 2,751 multi-location brands and roughly 350,000 locations, found that ChatGPT recommended just 1.2% of locations, compared with 35.9% visibility in Google's local 3-pack. In retail, only about 45% of the brands most visible in traditional local search overlapped with the brands most visible in AI recommendations. Strong SEO still helps, but it clearly does not guarantee AI search visibility.

• Manual AI visibility audits work best as a first diagnosis, not a long-term tracking system.
• The highest-leverage checks are crawl access, entity clarity, factual consistency, and evidence.
• Generic copy is a hidden AI problem because models struggle to extract and trust it.
• Structured data helps, but it does not compensate for vague pages or conflicting facts.
• The fastest reality check is to ask major AI engines the same factual questions and compare the answers.

Why does AI visibility need its own audit?

AI visibility needs its own audit because ranking signals and answer-generation signals overlap without being identical. A page can rank well in Google, yet still be hard for ChatGPT, Claude, Perplexity, or Google's AI experiences to summarize, compare, or recommend.

Traditional SEO audits look at crawlability, performance, backlinks, and on-page basics. Those still matter. But AI systems also care about whether your business is a clear entity, whether your claims can be corroborated, and whether your content contains reusable facts rather than polished but empty copy.

A simple example makes the gap obvious. A hotel can rank for "boutique hotel in Lisbon" because its domain is strong and its location page is optimized. If the site hides room policies behind tabs, uses generic descriptions for every room type, and shows different check-in hours across its own pages and listings, an AI system may still avoid recommending it.

That is why a real AI visibility audit is less about keyword placement and more about whether a machine can find, understand, trust, and restate your brand without confusion.

What are the 9 checks, in one view?

The nine checks below form a practical manual audit. They are not advanced model diagnostics. They are the fastest way to find whether your biggest problem is access, clarity, evidence, structure, or consistency.

If you run these checks in order, you avoid a common mistake: rewriting content before confirming the site is even reachable and the brand is even machine-readable.

Can AI systems actually reach and parse your site?

Start here because nothing else matters if the page is inaccessible. If AI crawlers cannot fetch your content, your beautiful copy and perfect schema are irrelevant.

Check your robots.txt file for directives that block GPTBot, ClaudeBot, PerplexityBot, or similar agents. Then look for practical barriers on the page itself: language selectors, location pickers, cookie walls, or JavaScript-heavy interstitials that appear before any real content. These are easy for teams to ignore because Google's crawler often still gets through, while newer AI bots may not.

A quick example: a travel site may expose destination pages only after a geo-selector popup. A human can click through. A crawler without that interaction path may never see the body copy at all. In that case, the problem is not ranking. It is basic eligibility.

If you want a simple retrieval test, BotRank recently explained how to test whether AI search can retrieve your page. If you want a broader technical review at scale, BotRank's GEO technical audits are built to surface the access and parseability issues that standard SEO crawlers often miss.

This is also where an llms.txt file can help. It will not override a hard block in robots.txt, and it will not magically create citations. What it can do is give AI systems a clearer, curated map of the pages you most want them to understand.

Do you have a real entity inventory, or just marketing pages?

A strong AI footprint starts with a clear entity inventory. That means listing the distinct things an AI system may need to describe about your business, then checking whether each one has explicit, consistent facts attached to it.

This is the core of Entity SEO. Your brand is one entity, but it is rarely the only one that matters. Products, service lines, locations, founders, room types, certifications, policies, integrations, and support options can all influence how a model understands and recommends you.

The most common gap is not the homepage. It is the sub-entities. A SaaS company may have a crisp homepage but almost no structured detail on who the product is for, what systems it integrates with, or which pricing rules apply to which plan. A hotel may describe the property well but say almost nothing concrete about room categories, nearby landmarks, parking, or cancellation terms.

One of the most overlooked categories is location context. "Close to downtown" is weak. "A seven-minute walk from Union Station" is usable. "Five minutes from the convention center by tram" is even better because it gives a specific relationship AI can restate.

BotRank has written in depth about how to audit your AI entity footprint. It is a useful mindset shift because many AI visibility problems are not content volume problems. They are business-definition problems.

Is your content specific enough for AI to reuse?

AI systems cite and restate pages more easily when the language is concrete. If a sentence is vague for a reader, it is usually worse for a model.

Look for pronouns with fuzzy references, empty adjectives, and category claims that do not resolve into facts. Sentences like "we deliver exceptional results through tailored solutions" say almost nothing. A sentence like "our onboarding team migrates HubSpot, Salesforce, and Pipedrive data in under 10 business days for mid-market B2B teams" says a lot.

This is where information gain matters. If your page says what every competitor could also say, AI has little reason to prefer it as a source. Unique numbers, named processes, first-party findings, and explicit constraints make a page more reusable. A bakery that says "fresh pastries daily" blends into the category. A bakery that says "laminated dough is prepared in-house at 4 a.m. every day, and our pastry menu changes twice a week" gives AI something distinctive to work with.

The broader principle is simple: AI visibility starts before the prompt because the model is assembling trust from what it already recognizes and what it can cleanly extract. BotRank explored that idea in this piece on why AI visibility starts before the prompt and ends with citations.

There is a nuance here worth keeping. Highly specific copy helps for factual retrieval and citation. It does not guarantee recommendation. If the page is specific but the offer is weak, overpriced, or badly reviewed elsewhere, AI can still choose someone else.

Can every important claim be verified, and do your facts match everywhere?

Verification and consistency are where many brands lose trust quietly. Unsupported claims and conflicting facts force AI systems to lower confidence or look elsewhere.

Start with your superlatives. If a page says "best," "leading," "award-winning," or "the only," ask what supports that statement. If the support exists, make it explicit. Name the award. State the year. Link the certification page internally. Clarify the scope of the claim. If the support does not exist, delete or soften the language.

Then audit factual consistency across your website and major public profiles. Hours, phone numbers, pricing rules, service areas, cancellation policies, shipping timelines, and category labels should align. This sounds basic because it is basic. It is also one of the highest-return free fixes you can make.

Consider a clinic whose site says same-day appointments are available, while its profile on a directory says bookings require 24 hours' notice. A human may forgive that discrepancy. A model comparing sources may not. The result is often an evasive or incomplete answer rather than a confident recommendation.

For teams that want to understand not just whether they are cited but which pages and sources are shaping those answers, BotRank's Source Analysis is especially relevant. It helps validate whether the pages AI cites actually mention your brand clearly and whether the supporting evidence is working in your favor.

Do structured data and knowledge connections reinforce your brand?

Structured data helps machines disambiguate what your page is about. It is not a magic button, but it is still one of the clearest ways to tell a system, in plain machine-readable terms, who the brand is and how related entities connect.

On a practical level, you can inspect a page source and search for application/ld+json. If nothing appears, there is no visible JSON-LD on that page. If it does appear, the next question is quality, not mere presence. Does it describe the right entity type? Is it reasonably complete? Does it include sameAs references to recognized profiles when appropriate? Does it reflect the facts currently shown on the page?

Thin schema often creates false confidence. A homepage with a bare Organization block is better than nothing, but it will not compensate for vague service pages, missing product detail, or contradictory listings. Schema works best when it mirrors a site that is already clear.

Knowledge connections matter too. Search your brand name. Do you trigger a recognizable Knowledge Panel? Does your structured data connect to authoritative identity references where relevant? Have you used the same brand name, same short description, and same category labels across major profiles?

This is one reason manual audits should combine page inspection with off-site checks. AI systems do not understand your business from one page alone. They assemble a picture from the wider evidence web.

BotRank's Take

Manual audits are a smart place to start because they force you to inspect the actual ingredients of AI visibility instead of hiding behind one score. The limit is that manual testing becomes noisy fast. Answers change by model, by prompt wording, by browser state, and by time. A snapshot tells you where to look. It does not tell you whether the issue is persistent.

That is why the most useful next step after a manual audit is repeatability. BotRank's AI Visibility tracking turns your prompt set into a stable measurement panel across multiple models, while its Source Analysis helps you see which pages and citations are actually shaping the answer. In this context, the value is not a vanity score. It is separating one-off weirdness from a real pattern you should fix.

What should you ask ChatGPT, Claude, Perplexity, and Google right now?

The best live test is a set of factual prompts that a buyer or customer might really ask. Ask the same questions in a fresh or private browsing session, log the answers, and compare them against reality.

Use prompts that force the model to show whether it understands your entity, your offer, and your proof. For example:

• What does [brand] do, and who is it for?
• Compare [brand] with two named alternatives for [specific use case].
• What are [brand]'s pricing or policy options?
• Is [brand] a good fit for [audience segment or company size]?
• What sources would you use to evaluate [brand]?

Score each answer manually on a few basics: were you mentioned, were you recommended, were you cited, and were the facts right? That distinction matters. A mention is not a recommendation. A citation is not proof that the answer was favorable. BotRank's readers will recognize this as the difference between raw appearance and meaningful visibility.

If you are doing this more than once, save the prompt set. That is exactly why reusable prompts matter. A good audit is not just a one-time poke at a model. It is a baseline you can repeat after content, technical, or profile changes.

How do you turn a manual audit into a real GEO action plan?

Turn the findings into a short, ranked backlog. Start with blockers that affect eligibility, then move to fixes that improve clarity and trust.

• Fix access blockers first. Remove unnecessary interstitial barriers and review bot rules.
• Standardize core facts. Align your site, profiles, directories, and support docs.
• Upgrade weak pages. Replace vague copy with named entities, explicit facts, and original evidence.
• Improve structured data. Expand thin schema so it matches the business you actually describe.
• Track the prompts again. Re-run the same questions after each wave of changes.

This is where a recommendation layer becomes useful. Once you know the problem, the hard part is turning it into work that someone can actually ship. BotRank's Recommendations feature is built for exactly that transition from diagnosis to execution.

The bigger lesson is simple. A free manual audit is enough to reveal whether your issue is crawl access, entity clarity, content specificity, claim credibility, or factual drift. That is already far more actionable than guessing why a model is overlooking you.

FAQ

Can I audit AI visibility without paid software?

Yes. A browser, a free AI assistant, and a repeatable checklist are enough to diagnose the biggest issues. The tradeoff is that manual checks are slower and harder to trend over time.

Which check should I run first?

Start with access and live-answer testing. If bots cannot reach your pages, or if major models already return wrong answers about your brand, you have immediate priorities.

Does schema markup guarantee AI citations?

No. Schema helps machines interpret entities and page meaning more clearly, but it does not replace specific content, proof, or off-site trust signals. Think of it as reinforcement, not a shortcut.

How often should I repeat a manual AI visibility audit?

Quarterly is a practical minimum for most brands, and monthly is better for competitive or fast-changing categories. Re-run the same prompts after major page edits, policy changes, or profile cleanups.

What is the difference between being mentioned and being recommended?

A mention means the model named your brand. A recommendation means it actively presented your brand as a good option for the user's need, which is a much higher bar.

If you want a no-cost starting point, run the nine checks above and document the gaps. If you want to make that process measurable across prompts, models, and cited sources, BotRank is the natural next step.