Can you make an old website understandable to ChatGPT and Perplexity without rebuilding it? Yes—by restructuring content, metadata, and entity signals while keeping the visual design intact. Most legacy sites fail AI readability not because of poor coding, but because their information architecture was designed for human scanning, not machine comprehension. This case note walks through how we approach such upgrades, what changes actually move the needle, and where the line between "optimization" and "redesign" sits.
Why legacy websites struggle with AI search engines
Modern AI search engines don't "crawl" pages the way Google does—they extract entities, relationships, and intents from your content. A legacy site typically has:
- Unstructured prose without clear headings, summaries, or Q&A blocks
- Missing schema markup (JSON-LD, RDFa) that helps AI map your offerings
- Navigation-driven copy (e.g., "Products › Solutions › About") rather than entity-driven copy
- Thin metadata—title tags and descriptions that read like keywords, not answers
The result: when ChatGPT or Perplexity tries to answer "who sells X in Y region," your legacy page either gets ignored or cited with wrong context. The fix isn't a new homepage—it's a systematic upgrade of how your content communicates with machines.
Our upgrade process (without touching the design)
We follow a four-step workflow that fits into your existing CMS or static site. No redesign, no new templates—just smarter content and structured data.
1. AI search health check (free)
We run a diagnostic on 5–10 key pages: checking schema presence, heading hierarchy, answerable question coverage, and entity mentions. This gives a baseline score and a prioritized fix list. Most clients discover that 60–80% of their AI visibility issues come from two or three systemic content patterns.
2. Content restructuring per page
We convert introductory paragraphs into a two-sentence "direct answer" block, then reorganize the body into clear H2/H3 sections with one idea per section. We add an FAQ block with 3–5 real questions customers ask. This doesn't change visual layout—it just makes the HTML structure more parseable.
3. Entity and schema enrichment
We add LocalBusiness, Product, Service, and FAQ schema (JSON-LD) to relevant pages. We ensure your brand name, product names, and service categories appear consistently with the same terms on every page—AI engines love consistency. We also add "mentions" of related entities (e.g., complementary services, competitor alternatives) to give AI more context.
4. Tracking and iteration
For businesses that want this managed on an ongoing basis rather than as a one-off fix, the Growth Engine (49,900 THB/mo) puts a specialist in charge month to month: they monitor how often your site appears in responses to a set of target queries across ChatGPT and Perplexity, and adjust content based on what they find—work that would otherwise mean hiring for the role internally. You see a before/after baseline within 2–4 weeks. If something isn't moving, we adjust the content—not the design.
What actually moves the needle? (Qualitative observations)
Based on client outcomes, the highest-impact changes are rarely "technical." They're:
- Answer-first intros: Pages that open with a crisp, direct answer to the most likely query (e.g., "We offer a free AI search health check so you can see where you stand") get cited more often than pages that start with marketing fluff.
- Explicit service lists: A simple bulleted list of services with one-sentence descriptions beats a paragraph of prose. AI engines extract list items with high confidence.
- FAQ blocks: Every page answers 3–5 questions that your sales team actually receives. This aligns with how AI training data is structured—questions and answers are natural entities.
- Consistent naming: If your homepage says "AI search optimization" but your pricing page says "GEO services," AI gets confused. We standardize terminology across the site.
We do not promise a specific ranking improvement—that depends on your niche, competition, and content depth. But most clients see a measurable increase in "mentioned as an option" responses within 4–6 weeks of implementation, based on their tracking dashboard.
DIY vs hiring a service
You can absolutely start on your own: add an FAQ section, rewrite your intro to an answer-first format, and install a schema markup plugin (free on WordPress). This will help—but it's like tidying a room vs. having a professional organizer. The DIY route takes 10–20 hours across multiple pages, risks missing cross-page entity consistency, and gives you no baseline data. If you have the budget and want a structured, measurable upgrade, our AI-Ready Website Retrofit (฿12,900) covers the free diagnostic, content restructuring, schema implementation, and monthly mention tracking setup—no redesign required.
FAQ
How long does a legacy site AI upgrade take without a redesign?
The active work typically takes 3–7 business days for a small-to-mid site (up to 20 pages) when done as a service. You'll see first tracking data within 2–4 weeks after implementation, as AI search engines' index refreshes are not instant. The design, layout, and user experience remain untouched.
Will this affect my Google rankings or site speed?
No negative impact is expected. The changes we make—adding schema, restructuring HTML headings, and enriching content—are all recognized by Google as positive signals. Because we don't add heavy scripts or change server configuration, site speed remains identical. The only "risk" is that your content becomes better organized, which can actually improve organic click-through rates.
What if my site is built on an old platform (e.g., legacy PHP, static HTML)?
The upgrade is platform-agnostic. JSON-LD schema and HTML restructuring can be added via a header/footer include or a small content management tweak. For static sites, we'll provide a simple snippet you paste into each page's `<head>` and body. If your platform blocks code editing entirely, we'll recommend a migration path in the health check—but that's a separate decision, not required for the optimization.