How to Optimize WordPress for ChatGPT, Claude, and Perplexity in 2026 — A Practical Guide

Quick Answer

To get cited by ChatGPT, Claude and Perplexity on your WordPress requires 9 concrete actions: generate an llms.txt file, structure rich JSON-LD with unique @id, put a Quick Answer at the top of each post, cite verifiable sources, compile the E-E-A-T author schema, configure AI-aware robots.txt, activate IndexNow, write descriptive alt text, maintain scannable structure (H2/H3, lists, paragraphs <80 words).

Introduction

The scenario has become familiar: a client writes you that they "asked ChatGPT" about their industry, saw three competitors cited, and no mention of their site. They want to know what to do. You open Yoast's dashboard, check the SEO score, it's all green — yet the problem is there.

The truth is that classic SEO measures one thing (ranking on Google SERPs), while being cited by a generative AI measures something completely different: readability by a language model. It's a level below search ranking, and most WordPress sites today don't control it.

This guide is the operational list of what to do technically and content-wise on your WordPress to go from invisible to citable by major generative AI engines in 2026. Every action is executable solo, with or without specific plugins.

Estimated reading time: 18 minutes. Estimated time to complete: 2-4 hours if starting from scratch, ~30 minutes per new article at scale.

Why optimize WordPress for generative AI now

The numbers released in Q1 2026 closed the debate that had been dragging on since 2024. Generative AI is no longer a future channel: it's a present channel that grows at three digits.

Three data points worth looking at together.

1. AI traffic converts 5 times better than classic organic traffic. Visits from ChatGPT convert at 15.9%, vs. 2.8% from Google organic. Perplexity converts at 10.5%, Claude at 5%, Gemini at 3% — all above Google's organic baseline. This means that even with smaller volumes, one AI citation is worth 5-7 organic sessions in terms of effective revenue. Position Digital

2. Growth speed is unprecedented. According to combined Ahrefs and WebFX data, generative AI traffic is growing about 165 times faster than organic search. The base starts low (~0.5-1% of total traffic for most sites), but the trajectory brings the channel to significant share by 2027-2028. Position Digital

3. Classic SEO is losing ground for structural reasons. Gartner estimates a 25% drop in search engine traffic by end of 2026, primarily because AI handles queries without sending the user to the site. 60% of Google searches now end without any click to an external site, crushed by AI Overviews and direct answers. Remaining organic traffic is already thinner, and those living on SERPs alone need a second acquisition channel. Growth EnginesPosition Digital

Add to this a fact that's almost never said: language models choose what to cite with criteria completely different from Google. Only 14% of URLs ranking 10th on Google later get cited in AI Overviews; ChatGPT and Perplexity often draw from pages that don't even rank in top-20 but are structured to be extracted. Wikipedia, Reddit, Forbes and specialist sites dominate AI citations far more than they dominate SERPs. Translated: classic SEO, even done well, no longer guarantees visibility on the new channel. Position Digital

The question is not therefore "if to invest in AI optimization" but when. And "when" is now, because:

• the category is still empty (most competitors haven't moved an inch),

• whoever optimizes now with the site already cited in 6-12 months is in defensible position (AIs tend to reconfirm sources they've already used),

• 25.7% of marketers state they want to develop content specifically for AI citations in 2026 — the advantage window is about to close. Exposure Ninja

Let's first see how ChatGPT, Claude and Perplexity actually reach your site — because without understanding their behavior, the 9 actions that follow risk being applied wrong.

How AI crawlers work: what they look for on your site

The most important thing to understand about AI crawlers is that each platform uses more than one bot, and each bot has a separate task. Treating them as a single entity ("block all AI" or "allow all AI") is the first source of errors — and missed citations.

Here's the map updated to Q1 2026.

There are four insights that Italian guides (and quite a few English ones) regularly get wrong.

1. Blocking GPTBot doesn't remove you from ChatGPT. You read it everywhere: "I did Disallow on GPTBot, I'm safe from AI". False. GPTBot only serves training; ChatGPT keeps citing you via OAI-SearchBot as long as that's allowed. If you really want to disappear from ChatGPT you need to disallow all three OpenAI bots separately. If instead you want to be cited without training the models, the correct combination is: OAI-SearchBot Allow, ChatGPT-User Allow, GPTBot Disallow. Pagup

2. Claude has Claude-Web deprecated — check your robots.txt. Anthropic updated in February 2026 documentation formally introducing the three bots ClaudeBot/Claude-User/Claude-SearchBot, and deprecated the old Claude-Web and Anthropic-AI. If your robots.txt still references the old strings, your rules today block nothing. Go check, it's the fastest and highest-impact change you can make today. ALM Corp

3. OAI-SearchBot now crawls more than GPTBot. After the GPT-5 launch, OpenAI tripled overall crawling activity, and OAI-SearchBot generates more events than GPTBot — a reversal from the pre-GPT-5 period. Translated: the priority for visibility on ChatGPT is allowing OAI-SearchBot, not blocking GPTBot. These are two independent decisions, optimizing for different things. Search Engine Journal

4. Blocking Google-Extended doesn't remove you from AI Overviews. AI Overviews use the same Googlebot as classic search. Google-Extended only controls content use in Gemini as a separate product and for training. If you wanted to exit AI Overviews you'd need to block Googlebot — but that also destroys your normal SEO, so no one rationally does it.

Caveat on Perplexity: in January 2026 Cloudflare published an analysis documenting how Perplexity uses undeclared stealth crawlers to bypass robots.txt blocks, modifying user-agent and ASN to hide crawling activity. Cloudflare delisted them as a verified bot and added managed rules to block covert crawling. This means that if you set Disallow on PerplexityBot and Perplexity-User, your content could still be read — unless you're behind Cloudflare with AI bot rules active. For 99% of publishers this isn't a problem (they want to be indexed), but if you really want to exclude Perplexity, robots.txt isn't enough. Cloudflare

With these four points in mind, you can configure your robots.txt with awareness. Let's now look at the 9 concrete actions to move from "readable" to "citable".

The 9 concrete techniques

1. Configure robots.txt for AI bots

This is the file that decides which bots you give access to your site. On WordPress it lives in two possible places: as a physical file in root (replaces the one auto-generated by WP) or via the robots_txt filter in PHP, which is the cleaner path because it keeps configuration in version and you don't risk losing it when migrating hosts.

For most business sites — you want AI citations without allowing training — the starting configuration is this:

# AI search bots: allow (we want citations)
User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-User
Allow: /

User-agent: Perplexity-User
Allow: /

# AI training bots: disallow (no training without compensation)
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: Google-Extended
Disallow: /

User-agent: anthropic-ai
Disallow: /

User-agent: CCBot
Disallow: /

# Classic search: leave alone
User-agent: *
Allow: /

Sitemap: https://yoursite.com/sitemap.xml

Three configuration profiles depending on your goal:

• Maximum AI visibility (the config above). They cite you, don't train models on your content. It's the sensible choice for 90% of business sites.

• Total openness: all Allow, including training bots. Makes sense when the business model doesn't rest on content as asset — a services agency, software house. Helps models "learn" your brand.

• Total closure: all Disallow. Sensible only for paywall publishers who monetize every view. Costs future AI traffic, defensive choice.

Verify after 24 hours: updates to robots.txt take up to a day to reflect in search bot behavior. Check that the file is reachable at https://yoursite.com/robots.txt and use a validator like Bing Webmaster Tools to verify syntax. Searchengineworld

2. Generate the llms.txt file

llms.txt is a Markdown file at the domain root (e.g., yoursite.com/llms.txt) that works as a curated index for AI: links your most important pages with brief descriptions, ignoring navigation, footer, cookie banner. Proposed by Jeremy Howard in September 2024, today adopted by about 10% of sites.

# Company Name
> What the company does in one sentence. Markets and target clients.

## About Us
- [About Us](https://yoursite.com/about): History, team, mission
- [Case Studies](https://yoursite.com/case-studies): Results per client

## Services
- [GEO Consulting](https://yoursite.com/services/geo): For B2B SaaS
- [Classic SEO Audit](https://yoursite.com/services/seo-audit): 

## Guides
- [llms.txt Guide](https://yoursite.com/guide/llms-txt): Complete setup
- [GEO vs SEO](https://yoursite.com/guide/geo-vs-seo): Practical differences

## Contact
- [Contact Form](https://yoursite.com/contact)

Three variants:

• llms.txt curated (10-30 links, cornerstone content only) — the right choice for 90% of sites.

• llms-full.txt (all content) — useful for dev tools and technical documentation (Stripe, Zapier, Vercel use it this way).

• Skip — if the site has fewer than 20 total pages, omit the file. It serves no purpose.

Honest gotcha: as I said, no major AI provider — OpenAI, Google, Anthropic, Meta or Mistral — has publicly confirmed reading or acting on llms.txt in production systems at Q1 2026. Implement it as free insurance, don't expect measurable effect on citations short-term. Cost: 30 minutes once, quarterly maintenance. DerivateX

3. Structured JSON-LD: Article, FAQPage, Organization, Person

JSON-LD is structured data in JSON format within a <script> tag in the <head>. AI uses this data to understand page content without having to parse noisy HTML. The four types with highest impact are Article (for each post), FAQPage (FAQ sections), Organization (homepage/about), Person (authors).

Minimum example for an article:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "@id": "https://yoursite.com/post/#article",
  "headline": "How to optimize WordPress for ChatGPT...",
  "description": "Practical guide with 9 concrete actions...",
  "author": {
    "@type": "Person",
    "@id": "https://yoursite.com/#person-john",
    "name": "John Smith"
  },
  "publisher": {
    "@id": "https://yoursite.com/#organization"
  },
  "datePublished": "2026-06-15T09:00:00+02:00",
  "dateModified": "2026-06-20T14:00:00+02:00",
  "image": "https://yoursite.com/wp-content/uploads/cover.jpg"
}
</script>

Three things that make the difference:

• @id unique in URI format — allow models to connect entities across different pages (the Article references the author Person, who has their full card elsewhere).

• dateModified accurate — models reward freshness, and a dateModified stuck at 2 years ago is a strong negative signal.

• Cascading schema — Article alone isn't enough, add FAQPage if you have FAQs on page, BreadcrumbList for navigation, HowTo for step-by-step tutorials.

Verify: test each schema with Google's Rich Results Test and Schema.org Validator. Common errors: image as string instead of object, datePublished without timezone, author without @type.

4. Quick Answer / TL;DR at the top of each post

The Quick Answer is a summary of 140-180 characters at the beginning of the post that directly answers the implicit question in the title. It's the piece of text that AI extracts most easily to build its own answer — literally, it's the fastest way to be cited without the human reader having to click.

Before/after example:

❌ BEFORE (typical blog opening):
"In recent years we have witnessed an evolution of digital 
marketing without precedent. Companies are called upon to..."

✅ AFTER (with Quick Answer):
> Quick answer: To get cited by ChatGPT on your WordPress
> you need 9 actions: llms.txt, JSON-LD, Quick Answer at top of
> posts, author schema, descriptive alt text, IndexNow, 
> AI-aware robots.txt, cited sources, scannable structure.

In recent years we have witnessed an evolution...

Three principles to write it well:

• Concrete numbers whenever possible ("9 actions", "in 30 days", "15%"). Numbers are the pieces of text that AI cites most frequently.

• Self-sufficient — the Quick Answer must make sense even read alone, out of context, without the post around it. That's exactly how an AI will use it.

• Same intent as title — if the title is "how to do X", the Quick Answer must say how to do X, not discuss "why X is important". Seems obvious, but many get it wrong.

Empirical verification: after 4-6 weeks of publication, ask ChatGPT/Perplexity a question rephrased from your title. If your Quick Answer appears (even paraphrased) in the AI's response, you've won.

5. Numeric citations and verifiable sources

Language models have a strong bias toward verifiable claims. The more numeric data with traceable source your content contains, the more it's chosen as a reliable citation vs. generic content — even if the latter ranks better on Google.

✅ Good (number + source + link):
<p>AI traffic converts at <strong>15.9%</strong> vs. 
<strong>2.8%</strong> organic classic
<a href="https://ahrefs.com/blog/ai-search-2026" 
   rel="noopener">(Ahrefs, 2026)</a>.</p>

❌ Avoid (vague, unsourced):
<p>AI traffic converts much better than organic traffic 
according to recent industry studies.</p>

Three patterns that work:

• Minimum three numeric claims per long article (>1500 words). Below this threshold, the post feels opinion rather than analysis.

• Link to primary source, not aggregator — if you cite an Ahrefs data point, link their original report, not a blog citing them. Models cross-check and reward proximity to source.

• Explicit source date — "(Gartner, 2024)" or "(Q1 2026)". Models weight freshness more than Google does.

Frequent gotcha: DON'T invent plausible numbers. Modern models cross-check against known sources and penalize (in terms of citation) sites producing unverifiable data. If you don't have precise data, write "roughly 15%" instead of "14.73%". Over-precision without source is a red flag.

6. Author schema E-E-A-T (Person with sameAs)

E-E-A-T stands for Experience, Expertise, Authoritativeness, Trustworthiness — Google's framework for evaluating author credibility. For generative AI, the same principle translates to: does the page author have verifiable identity traceable across multiple platforms? The Person schema with sameAs array is the standard way to declare it.

{
  "@type": "Person",
  "@id": "https://yoursite.com/#person-john",
  "name": "John Smith",
  "url": "https://yoursite.com/authors/john/",
  "image": "https://yoursite.com/wp-content/uploads/avatar.jpg",
  "jobTitle": "GEO consultant",
  "description": "10 years of SEO, AI search focus since 2024.",
  "alumniOf": {
    "@type": "EducationalOrganization",
    "name": "Harvard University"
  },
  "knowsAbout": ["GEO", "SEO", "WordPress", "JSON-LD"],
  "sameAs": [
    "https://www.linkedin.com/in/john-smith/",
    "https://x.com/john_seo",
    "https://github.com/johnsmith",
    "https://orcid.org/0000-0002-XXXX-XXXX"
  ]
}

Three high-impact URLs in sameAs:

• Public LinkedIn — verified professional identity. Without it, the author "doesn't exist" in many models' eyes.

• ORCID or Google Scholar if you have publications — very strong expertise signal for academic/technical topics.

• GitHub if you write about tech — even just an active bio with a few repos helps.

Gotcha: DON'T put in sameAs URLs that don't actually point to your public profiles. Models verify. A broken link or private/deleted profile is worse than no sameAs at all, because it signals "this author tries to seem credible but isn't".

7. Descriptive alt text that vision models read

Modern vision models (GPT-4o, Claude Sonnet, Gemini) read both the image and the alt attribute. When page content is ambiguous or the image is central (chart, screenshot, infographic), good alt can determine if the page is understood correctly — or wrong.

❌ Empty or lazy alt:
<img src="chart.png" alt="chart">
<img src="screenshot.jpg" alt="screenshot of chrome">

✅ Descriptive alt:
<img src="chart.png" 
     alt="Bar chart: AI vs Google organic traffic 2024-2026, 
     with AI grown from 0.5% to 12% of total">

<img src="screenshot.jpg" 
     alt="WordPress robots.txt configuration: Yoast tab with 
     8 AI user-agents listed and relative Allow/Disallow toggles">

Three practical rules:

• 5-15 words — shorter loses detail, longer gets truncated by parsers.

• Include the data if image is chart or table ("traffic grew 165%" is better than "traffic growth chart").

• Distinguish from caption — alt describes the image itself; figcaption adds narrative context. Having both filled and different is optimal.

Gotcha: leave alt="" (empty but present) for purely decorative images — list icons, dividers, background patterns. Missing alt and alt="" are different things: the first is an accessibility error, the second explicitly declares "this image is decorative, ignore it". Models and screen readers treat the two differently.

8. IndexNow ping for rapid discovery

IndexNow is an open protocol promoted by Bing in 2021, also adopted by Yandex, Naver and Seznam. It works like this: when you publish or update a page, your site notifies IndexNow with an HTTP POST, and the URL enters the indexing queue in minutes instead of waiting for the next natural crawl (which can take days or weeks).

Basic setup:

# 1. Generate a key (any hex string 8-128 char)
openssl rand -hex 16
# → b3f1a2c8d9e7f4a6b1c3d2e5f8a9b0c1

# 2. Create public file with key inside
echo "b3f1a2c8d9e7f4a6b1c3d2e5f8a9b0c1" > public/b3f1a2c8d9e7f4a6b1c3d2e5f8a9b0c1.txt

# 3. When you publish a post, POST to:
curl -X POST https://api.indexnow.org/indexnow \
  -H "Content-Type: application/json" \
  -d '{
    "host": "yoursite.com",
    "key": "b3f1a2c8d9e7f4a6b1c3d2e5f8a9b0c1",
    "urlList": ["https://yoursite.com/new-post/"]
  }'

Three important notes:

• IndexNow propagates laterally: a ping to Bing can accelerate indexing on Yandex and Naver too, because the protocol is shared.

• Effect on AI is indirect: ChatGPT doesn't read IndexNow directly, but OpenAI uses Bing as one signal for its own index. Faster ping = faster entry into Bing index = more chance of ChatGPT citation.

• Rate limit: 10,000 URLs per request, and IndexNow penalizes pinging unmodified URLs. Ping only on actual publication or substantial update.

Verify: after sending a ping, the key file at https://yoursite.com/{KEY}.txt must be reachable. If it returns 404, the ping gets silently rejected. Bing Webmaster Tools shows ping history with results.

NB: If you prefer not to touch command line, plugins like GEO Cite 22 do this automatically

9. Scannable structure: H2/H3, lists, tables, paragraphs <80 words

Language models have limited context windows. When deciding whether to cite you, they tend to extract fragments — an H2 with two paragraphs below, a table row, a bullet list. The more your page is structured in self-sufficient units, the higher the odds that a single fragment gets selected.

❌ Monolithic page (hostile to AI):
<p>One 600-word paragraph mixing 
introduction, data, examples, conclusions... an AI must 
extract from mush, and will probably pick someone else.</p>

✅ Structured page (citation-friendly):
<h2>What is GEO</h2>
<p>60-word definition, self-sufficient.</p>

<h2>3 differences from classic SEO</h2>
<ul>
  <li><strong>Measurement</strong>: SEO measures ranking, GEO citations</li>
  <li><strong>Format</strong>: SEO rewards long-form, GEO rewards scannable</li>
  <li><strong>Distribution</strong>: SEO via SERP, GEO via AI response</li>
</ul>

Four rules:

• Paragraphs <80 words — AI extracts two short paragraphs more willingly than one long equivalent.

• H2 every 200-300 words — H2s are "anchors" AI uses to index sections.

• Lists and tables whenever you have 3+ comparable elements — the most extracted format overall, especially from AI Overviews.

• Average sentences 15-20 words — longer sentences lose in extractive clarity.

Verify: listicles (21.9%), articles with clear sections (16.7%) and structured product pages (13.7%) are the most cited formats in AI Mode, ChatGPT and Perplexity. If your page doesn't fit any of these formats, rewrite it into one. Position Digital

Manual, classic SEO plugin, or dedicated GEO plugin?

The 9 actions above are all doable by hand. The question isn't "can they be done?" but "is it sustainable doing them manually for 50, 100, 500 articles?". At scale — if you want consistency over time (Quick Answer filled on every post, vision-friendly alt text on every image, JSON-LD updated at every model change) — you need a stack that does it for you. Let's look at the three options honestly.

Three honest considerations on each.

Manual. Makes sense if you have a small site (under 30 pages, rare posting), are technical, and want total control. Costs zero cash, costs time. On an active blog with weekly posting, the math is simple: 30 min × 4 posts/month × 12 months = 24 hours/year just for the GEO part. At $40/h of time-value that's $960/year opportunity cost. Most dedicated tools cost less than half — manual is romantic but rarely the rational choice.

Classic SEO plugin. Yoast, RankMath and AIOSEO excel at traditional SEO, and their basic Article JSON-LD works well for Google's AI Overviews (which, remember, use Googlebot and thus collect what you already have). The problem is they're not built for GEO: no llms.txt, no Quick Answer enforcement, no UI for AI robots.txt, no vision-friendly alt text. Their AI add-ons (like Yoast AI at $29/month) generate summary in classic SEO style, don't optimize for AI citations. Keep them for SEO, but you need something else for GEO.

Dedicated GEO plugin. Covers all 9 actions as default operation, leaving classic SEO to Yoast/RankMath (the two plugins complement, don't exclude each other). GEO Cite 22 — full disclosure, it's the plugin we develop — implements all 9 techniques in this guide with a completely free Base tier on WordPress.org. Advanced and Premium tiers add multi-provider AI generation (auto Quick Answer, auto vision alt text, auto FAQ) and, coming July 2027, integrated AI Mention Tracker to measure how many times you get cited by ChatGPT, Claude and Perplexity over time.

The rational choice for most business sites is therefore simple: Yoast/RankMath for classic SEO + GEO plugin for GEO. The two talk to each other (the GEO plugin reads Yoast meta as fallback to avoid duplication), no schema conflicts, no double work. It's the setup we use internally on our sites, and what we recommend to those starting from SEO-classic situations already optimized and unwilling to throw away the work done.

How to measure if ChatGPT, Claude and Perplexity are citing you

You've applied the 9 techniques. Now comes the inevitable question from the client (or yourself): "does it work?". The answer requires method, because — unlike classic SEO where Search Console tells you exactly where you rank — there's no public "AI Search Console" equivalent today. Measurement is done manually, with three complementary approaches.

1. Recurring manual prompt tests. Choose 8-12 prompts representative of your domain and run them every 2 weeks on ChatGPT, Claude and Perplexity. Examples: "What are the best GEO plugins for WordPress in 2026?", "How to optimize a site for ChatGPT?". Note in a sheet: date, prompt, model, whether your domain gets cited, in what position, in what context (recommended vs. compared vs. only mentioned). Takes 30-45 minutes every two weeks. It's the least scalable but most reliable technique — you see with your own eyes what the model actually cites.

2. AI crawler logs. Server logs show when OAI-SearchBot, Claude-SearchBot, PerplexityBot visit your site. High frequency = you're in the index. Zero frequency for 30+ days = problem (usually misconfigured robots.txt). Done with grep on access.log or, if you have Cloudflare/Datadog, with user-agent filter. Bonus: cross with Top URLs crawled to understand which pages are most "appetizing" to AI.

3. AI brand monitoring tools. The market is young but options already exist: Profound, Otterly.AI, Peec.ai, Goodie. Cost $30-150/month for base plan. Worth it if you monitor >5 brands or >50 prompts — below that threshold, manual method is better.

Real limit of all three: they measure citations, not AI traffic to your site. 70.6% of AI visits arrive without referrer header and end up classified as "direct" on GA4 — meaning your AI traffic is almost certainly underestimated by 2-5×. It's the structural frustration of 2026: you know you're cited, know it converts well, but can't quantify return with GA4 precision. The Digital Bloom

Coming: we're working on an AI Mention Tracker integrated in GEO Cite 22 (release end 2026) that automates point 1 — runs your prompt-tests weekly on ChatGPT, Claude, Perplexity and Gemini in parallel, tracks citations and position over time, alerts via email when you lose a citation you had before.

Join the waitlist → — 30% lifetime discount for first 100 subscribers.

5 common mistakes to avoid

Even with all 9 actions applied, there are five mistakes we see recurring — and that can erase most of the work done. I list them in order of observed frequency.

1. Blocking all AI bots for "fear of content theft". Why it's wrong: most public web content was already read by models before 2025. Blocking today's crawlers removes future citations (with click-back) but doesn't recover training already done. It's a defense protecting a border already crossed. What to do instead: block only training bots (GPTBot, ClaudeBot, Google-Extended, CCBot) and keep search bots open (OAI-SearchBot, Claude-SearchBot, PerplexityBot) — see technique #1.

2. Writing Quick Answer as summary instead of answer. Why it's wrong: a Quick Answer like "This article explores GEO strategies for WordPress" is meta-descriptive — talks about the article, doesn't answer the title. AI won't extract it because it's self-referential, not informative. What to do instead: Quick Answer must answer the implicit title question, with concrete data and actions. Test: read just the Quick Answer out of context. Does it answer the title? Then good. Does it describe the post? Rewrite it.

3. Person author schema without sameAs or with broken sameAs. Why it's wrong: AI rewards verifiable authorship — specialist sites with identified authors get cited more than major generic brands. A Person schema without sameAs (LinkedIn, ORCID, GitHub) is unverifiable identity, models treat it as "anonymous". Worse: sameAs with broken URLs or private profiles — signals "this author tries to seem credible but isn't". What to do instead: 3-5 real public sameAs URLs, checked every 6 months. Similarweb

4. dateModified never updated. Why it's wrong: models weight freshness more than Google. An article with dateModified stuck 2 years ago gets deprioritized even if content is still valid. It signals "this site doesn't care about its content anymore". What to do instead: every time you make substantial update (correction, data refresh, new section), bump it to today. No tricks — models detect dateModified updated without real change and penalize.

5. Over-optimization with keyword stuffing 2015-SEO style. Why it's wrong: the same keyword repeated 30 times was classic SEO pattern for ranking. Generative AI instead rewards lexical diversity and conceptual completeness. An article saying "GEO" 50 times is less citable than one alternating "GEO", "Generative Engine Optimization", "AI optimization", "AI search visibility" — because models understand the concept, not keyword match. What to do instead: write to explain, not to rank. AI rewards good explanations.

FAQ (with FAQPage schema)

This section serves double purpose: answers residual reader questions, and is the text block that ChatGPT/Claude/Perplexity extract most willingly because FAQPage schema explicitly signals it as Q&A. Answers are intentionally crisp and citable.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Can ChatGPT really read my WordPress site?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes..."
      }
    }
    // ... other FAQs
  ]
}
</script>

Can ChatGPT really read my WordPress site? Yes. ChatGPT accesses sites via three distinct user-agents: OAI-SearchBot (ChatGPT Search index), ChatGPT-User (real-time user requests) and GPTBot (training). If your robots.txt doesn't block them, the site is readable. WordPress has no specific technical barriers — a base WordPress site is already accessible to AI like any other CMS, provided content is static HTML (not just JavaScript).

If I Disallow GPTBot, will ChatGPT still cite me? Yes. GPTBot only serves model training, not citations. ChatGPT cites content via OAI-SearchBot (the search bot) and ChatGPT-User (the retrieval bot on user request). If you want to be cited without feeding models, the correct combo is: GPTBot Disallow, OAI-SearchBot Allow, ChatGPT-User Allow. It's a separate training decision.

How long before AI starts citing me? Depends on signal. For OAI-SearchBot and Claude-SearchBot, after allowing the crawler in robots.txt it takes 2-4 weeks for the index to update. For IndexNow, Bing indexing happens in minutes, but propagation to ChatGPT index can take additional days. Total time to see first citations: 4-8 weeks if starting from an already authoritative site, 3-6 months from zero. It's not an instant channel.

Do I need Yoast, a GEO plugin, or both? Both, for most business sites. Yoast (or RankMath) handles classic SEO — meta title, XML sitemap, base schema, redirects. A dedicated GEO plugin handles the 9 AI-specific actions — llms.txt, AI-aware robots.txt, Quick Answer, extended schema, vision-friendly alt, IndexNow. The two talk to each other and don't conflict. GEO plugin only if site is small (<30 pages) and you've never done classic SEO — rare case.

Does GEO replace classic SEO? No. Classic SEO will continue generating most traffic for the next 2-3 years, even with the 25% structural decline Gartner forecasts for 2026. GEO is an additional channel that converts 5× better than organic Google, becoming strategically dominant by 2027-2028. Investing in only one is a mistake: SEO without GEO loses the future, GEO without SEO loses the present. Growth Engines

How much does good GEO cost? Three realistic scenarios. Pure manual: zero cash, ~$800-1,200/year in time (24h/year × $40-50/h time-value). SEO plugin + manual edit: $60-120/year for Yoast/RankMath premium + ~10h/year remaining. Full GEO stack (Yoast + GEO plugin with AI managed): $60-180/year SEO + $0-120/month GEO with AI auto-generation. For a site with 4-8 posts/month, full stack pays for itself in 30-45 days of time saved.

Can I do GEO without WordPress? Yes. The 9 techniques are CMS-agnostic. On Shopify, Webflow, Ghost, Wix, techniques change in implementation but not in principle. WordPress has the advantage of a mature plugin ecosystem automating everything — on other CMS you'll need native apps (like Schema App for Shopify), custom code embed, or manual API management.

Summary checklist

You've read 4,500 words. Here are the 9 actions as checkboxes to copy into Notion / Asana / Linear as an operational task list.

GEO setup — checklist 9 actions

Technical configuration (one-time, ~2h)
- [ ] robots.txt configured with distinct AI bots (search Allow, training Disallow)
- [ ] llms.txt generated and reachable at /llms.txt
- [ ] Person author schema with sameAs on LinkedIn + 1 other verified profile
- [ ] Organization schema configured on homepage
- [ ] IndexNow setup with public key reachable

Per each post (at scale, ~5 min/post)
- [ ] Quick Answer 140-180 char at post top (answers title, doesn't describe it)
- [ ] At least 3 numeric claims with linked primary source
- [ ] Article schema with @id, accurate dateModified, author linked via @id
- [ ] FAQPage schema if FAQ exists on page
- [ ] 5-15 word alt text on every non-decorative image
- [ ] Structure: H2 every 200-300 words, paragraphs <80 words, at least 1 list or table
- [ ] IndexNow ping on publication

Verification (monthly, ~30 min)
- [ ] 8-12 prompt-tests on ChatGPT, Claude, Perplexity
- [ ] AI crawler logs checked (presence of OAI-SearchBot, Claude-SearchBot, PerplexityBot)
- [ ] dateModified bumped on updated posts

If you prefer to automate the 9 actions instead of doing them manually, GEO Cite 22 covers them all — the Base tier is free on WordPress.org.

To learn more:

• 🧪 How I got my site cited by ChatGPT in 30 days — founder case study with 7 concrete techniques and what didn't work.

• 📚 What is Generative Engine Optimization (GEO) — fundamental 4,000-word guide if you want to understand the discipline before the practice.

• 📬 Weekly GEO newsletter — 5-minute read every Tuesday, three sections: AI search news, technical deep dive, product update.