A 10-Point GEO Audit for Your Website

Can anything actually read your site

1. Is the site actually crawlable. Before anything about positioning or content quality matters, check whether a retrieval bot can even get the text off your pages. If your core content — the homepage pitch, product pages, docs — only renders after JavaScript runs, a lot of retrieval and indexing passes will see a mostly empty shell. Load your key pages with JavaScript disabled, or view source and look for the actual copy sitting in the HTML, not just a div waiting to be filled in by a script.

Also check robots.txt and any bot-blocking rules for AI crawlers like GPTBot, ClaudeBot, and PerplexityBot. Some sites block these by default through a security plugin, a CDN "bot fight mode," or a WAF rule that someone turned on to stop scraping and never revisited, with nobody deciding on purpose to shut out the crawlers that matter now. It is worth checking server logs too, if you have access to them, to see whether known AI crawler user agents are actually hitting your pages and getting a 200, rather than assuming your robots.txt is the whole story.

Pass: your core pages render as readable HTML without JavaScript, and you have confirmed — not assumed — that you are not blocking the crawlers you want to be found by. Needs work: content is client-rendered only, your robots.txt disallows the very bots that would surface you in an answer, or you simply have not checked either one in the last year.

Do you say what you do, clearly and consistently

2. Is there a clear one-paragraph statement of what the product does on the homepage. Somewhere near the top of your homepage, in plain language, there should be a paragraph that says who the product is for and what problem it solves — not a headline built from adjectives. This is the sentence you want a model to effectively paraphrase back when someone asks about you. Pass: you can point to one paragraph and say "that's the description." Needs work: the closest thing you have is a tagline plus a hero image, and the actual explanation is scattered across three sections a reader has to piece together.

3. Does every core page have a direct-answer structure. Pricing, features, comparison, and FAQ pages should each open with something close to a question as the heading and the answer in the first sentence or two beneath it, before any elaboration. This is the format that both retrieval systems and human skimmers can lift cleanly. Pass: the first paragraph under most headings on your key pages could be read on its own and make sense. Needs work: pages open with scene-setting, story, or a slogan before getting to the actual point.

4. Is positioning consistent across every page. Read your homepage, about page, pricing page, and any recent landing pages back to back and check whether the core claims agree — same category, same target user, same core problem. Drift happens naturally as different pages get written by different people at different times, and a model has no way to know which version is current, so it either hedges or picks the version it saw most often. Pass: the "what this is" sentence is functionally the same everywhere you look. Needs work: your homepage calls it one thing, your pricing page implies something broader, and a landing page from a campaign six months ago describes a different product entirely.

Does anyone else back up what you say

5. Does the site have genuine third-party mentions to point to. A model weighs independent confirmation heavily — it trusts a claim more when it isn't only coming from you. Check whether you actually have reviews, mentions in other people's roundups, forum threads where real users describe you accurately, or press coverage, and whether any of it is visible or linked from your own site. Pass: you can list several sources, written by people who are not you, that describe your product in terms roughly matching your own pitch. Needs work: the only descriptions of your product that exist anywhere are the ones you wrote yourself.

6. Are there comparison or alternatives pages that are honest rather than only self-flattering. "X vs Y" and "alternatives to X" content is exactly the shape of material models draw on when someone asks a comparison question, but a comparison page that only ever concludes you win reads as marketing and gets discounted accordingly, by readers and by models trained on a lot of marketing copy. The useful version names real competitors by name, describes what they actually do well, and is specific about who each option fits rather than insisting everyone should pick you.

Pass: your comparison pages name real competitors accurately, concede real tradeoffs, and would not embarrass you if a competitor's customer read them. Needs work: you have no comparison content at all, or what exists is thinly disguised self-promotion where every category somehow favors you.

Is your content built to be extracted and trusted

7. Is there a real FAQ section structured for extraction. A genuine FAQ — actual questions people ask, each with a direct answer in a sentence or two before any extra detail — is one of the easiest formats for a model to lift cleanly and attribute to you. Pass: your FAQ questions read like things a prospect actually typed into a search bar or asked in a sales call, and each answer stands alone. Needs work: the FAQ is decorative, restating your feature list as questions nobody would ask.

8. Are claims backed by specifics rather than adjectives. "Powerful," "seamless," and "best-in-class" carry no information a model can verify or repeat with confidence. Numbers, named integrations, concrete workflows, and specific outcomes do. Pass: when you scan your site for adjectives, most of them are attached to a specific fact nearby. Needs work: your copy is heavy on superlatives and light on anything a skeptical reader could check.

9. Is content fresh and updated rather than stale. Pages with dead links, outdated screenshots, pricing that no longer matches reality, or a "last updated" date from two years ago signal to both readers and retrieval systems that the page may not reflect current reality. Retrieval passes tend to favor pages that look actively maintained when there are multiple sources to choose between, and a page that visibly contradicts your current pricing or feature set is worse than no page at all, since it actively misinforms whoever reads it.

Pass: your key pages have been reviewed and touched within the last few months, and nothing on them visibly contradicts how the product works today. Needs work: you have pages ranking or getting traffic that describe a version of the product that no longer exists, or the newest date anywhere on your site is from a launch two years ago.

What do the models actually say about you right now

10. Can you verify what AI models currently say about you as a baseline. Before you change anything, ask ChatGPT, Claude, and Perplexity the actual questions a prospect would ask — "what's a good tool for X," "alternatives to Y," "does anything integrate with Z" — and note whether you're mentioned, what's said about you when you are, and whether it's accurate. Write the answers down somewhere, with the date, so you have something to compare against later rather than relying on memory or a vague impression from one time you checked.

This is the only way to know if the first nine checks are working, and it is the one step people skip because it takes real effort to do properly and repeatedly across multiple models and multiple questions. Pass: you have a written record of what several models say about you today, checked within the last month, covering the handful of questions your actual prospects would ask. Needs work: you are guessing, or you checked once a year ago on a single model and have not looked since.

Once you have run through all ten, you will usually end up with a short, specific list rather than a vague sense of "we should do more GEO." Fix the technical blockers first, since nothing else matters if a crawler cannot read the page. Then tighten positioning so it is consistent everywhere. Then go build the corroboration and content pieces you are missing — comparison pages, an honest FAQ, third-party mentions — since those tend to be the slowest to accumulate and the easiest to keep neglecting. This is also the part of the work that is hardest to do alone on top of everything else running a startup requires, which is one of the reasons Wally exists: it can research where your category gets discussed, draft the comparison and FAQ content the audit turns up as missing, and keep an eye on positioning consistency across new pages as you publish them, with everything queued for your approval before it goes out.