The homepage is the page AI bots read most on your site. Those bots don't look at the screen: they read the text in the HTML. Anything on that page that proves your value only as an image (a badge, a star rating, a row of client logos, a figure inside an infographic) escapes them.
This is where two disciplines that rarely spoke now meet. CRO, conversion rate optimization, works the page for the human who lands on it. GEO works it for the machine that decides whether that human lands at all. On the homepage, it is now the same job.
Andy Crestodina, co-founder of Orbit Media, put it this way in a conversation with Ross Hudgens, founder of Siege Media:
"Why would a machine trust this? It doesn't render the page, and look at the pixels. So I think if proof points are purely visual, then we need to bust them out and put them there also as text."
Why does the homepage matter so much to AI?
Because it is where engines come to learn who you are. Orbit Media analyzed more than 560,000 AI bot requests across 74 websites, using Cloudflare's AI Crawl Control reports. Bots requested the homepage about 15 times more than other page types.
The same study adds two lessons:
- Nearly half of all pages earn nothing. 47% of the pages studied generated no visit from an AI engine.
- Site size matters. The total number of URLs correlates strongly with the attention AI bots give a site.
On a separate sample of 27 websites, Orbit Media also finds that the homepage receives by far the most visits sent by AI, well ahead of service and product pages.
Crestodina's conclusion is pragmatic: if you only rework one page, make it this one. And the height of the page is not a problem. A long homepage that says everything an engine needs to know beats an elegant one that merely hints at it.
What does an AI bot actually see on your page?
Text, and only the text in the HTML the server returns.
Vercel and MERJ observed AI bot behavior across Vercel's network. None of the major bots tested runs JavaScript: not GPTBot or OpenAI's other bots, not ClaudeBot, not PerplexityBot. They sometimes fetch JavaScript files, but don't execute them. The notable exceptions are Gemini, which goes through Googlebot's infrastructure, and AppleBot.
Two direct consequences for a homepage:
- Script-loaded content doesn't exist. An animated counter, a testimonial carousel loaded in the browser, a review block injected by a third-party widget: if the text isn't in the source HTML, most AI engines don't read it.
- An image is not a fact. Bots do download images, and ClaudeBot spends a large share of its requests on them. But an award, a rating or a figure that only exists in pixels is not read as a claim made by the brand. At best, the engine reads the alt text.
A purely visual proof point is a conversion signal that only exists as an image, an animation or a script. The human sees it. The AI engine doesn't.
Which proof points should be written out as text?
Every one that matters in the decision. The principle is simple: each proof point keeps its visual form for the human eye, and gains a written, visible version in the HTML.
| Conversion signal | What the human sees | What to write |
|---|---|---|
| Ranking or award | A badge | The name of the ranking, the year, what it recognizes |
| Customer reviews | Five stars | The rating, the number of reviews, the platform |
| Reference clients | A row of logos | A sentence naming them, with their industry |
| Key figure | An infographic | The figure as text, with its source and date |
| Certification | An icon | The standard, the scope, the issuing body |
| Case study | A thumbnail and a PDF | The quantified result in the headline |
Crestodina cites typical examples: "We're in the Gartner Magic Quadrant," "We won this," "We're the number one hospital for this." As a badge, those claims don't train the engine. Written out, they become facts it can reuse.
One important caveat: writing the proof as text doesn't mean writing it for bots only. Text visible to bots but hidden from visitors is cloaking, which Google explicitly penalizes. The written version has to be readable by everyone: a line under the badge, a caption, a sentence in the trust section.
Why don't your case studies count for AI?
Because the proof is buried. A "Case study: Alba & Co" thumbnail on the homepage, leading to an eleven-page PDF whose key result sits on page nine, tells the engine nothing. It would have to follow the link, open the document, read it in full and pull out the right figure.
The rule is the same as for a headline that converts: the result goes in the headline.
| Before | After |
|---|---|
| Case study: Alba & Co | Alba & Co cuts its acquisition cost by 18% in six weeks |
| How Maison Lin rethought its SEO | Maison Lin doubles organic traffic on its key categories in a year |
Illustrative examples, fictional brands.
The same headline works three times: it convinces the hurried visitor on the homepage, it gives the AI engine a dated, attributed fact, and it introduces the full page, which itself has to exist as HTML and not only as a PDF.
Are CRO and GEO the same discipline?
Not quite, but on the homepage they converge, around the same reflexes:
- An explicit value proposition. What you do, for whom, with what result, in one sentence. What reassures a visitor in the first five seconds is also what an engine summarizes best.
- Proof next to the claim. CRO knows a promise without nearby proof converts poorly. For an engine, a claim with no verifiable fact nearby is unlikely to be picked up.
- Clarity over effect. An animation that reveals a figure on scroll impresses the human and leaves the engine with nothing. The written figure does both.
The two disciplines diverge on one point: length. CRO long pushed toward short, visual homepages. GEO rewards a page that says everything. The answer is not to choose but to structure: a fast, visual first part for the human, then written sections, organized by subheadings, that an engine can work through.
For Crestodina, this is where conversion optimizers and SEOs have the most to learn from each other: the full-funnel marketer has an edge, because they think in systems.
How do you check what AI reads on your homepage?
Four tests, from simplest to most complete.
- Read the source code. Open the homepage, view the page source (not the inspector, which shows the page after scripts have run) and look for your proof points: the award, the rating, the client names, the figures. What isn't there doesn't exist for most AI engines.
- Turn JavaScript off. Reload the page with JavaScript disabled. Whatever disappears, counters, carousels, review widgets, is invisible to GPTBot and ClaudeBot.
- Read your server logs. Find the requests from GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot. You will see which pages they request and how often, very likely starting with the homepage.
- Ask the engines. Ask ChatGPT, Gemini and Perplexity which awards your brand has won, which clients it serves, what rating it holds. The gaps with reality show which proof points don't get through.
The fourth test is the one that is hard to run by hand, because it has to be repeated across several engines and over time. That is what Nanga's GEO module does. The AI audit, in beta, checks the engines' answers against your own source of truth, describes the image they give of your brand and verifies that their crawlers can reach your site. AI traffic then measures in GA4 the visits assistants send you: the homepage is the most-read page, and it pays to know what it brings in.
Written proof also changes the sources engines read: a homepage that states its facts clearly becomes a citable source, not just a visited page. The topic is covered in the article on the sources each AI engine reads, and checking what AI says about you in the guide to auditing AI accuracy.
What Nanga changed on its own homepage
Nanga ran these tests on its own homepage first, before publishing this article. The finding looked like that of many sites: the proof existed, but mostly in visual form.
| Proof point | Before | After |
|---|---|---|
| Reference clients | A row of logos, readable only through their alt text | A sentence naming them: Vestiaire Collective, Ray Studios, 24S, Lunii and Le Slip Français |
| Customer reviews | Absent from the homepage | "215 certified reviews on Trustfolio," as text, with a link to the platform |
| Number of clients | Absent from the homepage | "Over 500 clients served by the team behind Nanga" |
| Hosting and GDPR | Left to the FAQ, at the bottom of the page | Written at the top of the page, under the call-to-action buttons |
Nothing is reserved for bots: these lines show for every visitor, right under the buttons at the top of the page, and sit in the source code without waiting for a script to run. The logos stay in place, for the human eye.
And your homepage? The first test takes two minutes. Open its source code and look for your best proof point: the rating, a client's name, the figure that closes deals. If you can't find it, neither can AI engines.
To see what AI engines take away from your homepage: request a demo.
Frequently asked questions
Do AI bots read JavaScript?
Mostly not. According to the analysis by Vercel and MERJ, GPTBot, ClaudeBot and PerplexityBot fetch JavaScript files without running them. The notable exceptions are Gemini, which relies on Googlebot's infrastructure, and AppleBot. Content injected in the browser does not exist for most AI engines.
Why does the homepage matter so much to AI?
Because it is the page their bots request most. Across 560,000 AI bot requests analyzed by Orbit Media, bots requested the homepage about 15 times more than other page types. It is where an engine learns who you are.
Is a badge or a client logo enough for AI to know about an award?
No. An award that only exists as an image is not read as a fact. Write it out as text, in the page's HTML: the name of the ranking, the year, what it recognizes. The badge stays useful for the human eye, alongside it.
Should text be added for bots only?
No. Text visible to bots but hidden from visitors is cloaking, which Google penalizes. The written proof has to be visible to everyone: a line under the badge, a caption, a sentence in the trust section.
Are CRO and GEO compatible?
Yes, and they reinforce each other. Both reward an explicit value proposition, proof placed next to the claim and a quantified result in the headline. What convinces a hurried visitor is often what an AI engine extracts most easily.
Sources:
- Andy Crestodina, in conversation with Ross Hudgens (Siege Media), quoted by Ross Hudgens on LinkedIn, 2026.
- Orbit Media, AI Crawls vs. AI Traffic: What 560K+ Requests Reveal, 2026.
- Vercel and MERJ, The rise of the AI crawler, December 2024.
- Google Search Central, Spam policies: hidden text and links.