Your site has traffic. AI engines still ignore you. Here's why - and how to fix it.

Why do sites with Google traffic get ignored by AI engines?

A founder runs a CiteAgentic scan. The site is well-indexed. It has 5,000 organic visits a month from Google. The blog is updated. The product is real. The team is competent.

The visibility score comes back at 8%.

That gap - solid SEO, near-zero AEO - is the single most common pattern we see. It's not random. It maps to four specific failure modes, in this order of frequency:

1. The lead-answer is buried. The page answers the buyer's question, but the answer is in section 4. AI engines truncate at ~512 tokens during reranking. They never see your good stuff.
2. The brand name doesn't appear next to the answer. Even when the answer is up top, the brand isn't named alongside it. AI engines extract spans containing entities; if your span has no entity, the citation goes elsewhere.
3. The page has no Hearst patterns or sourced claims. The content is well-written prose without the structural signals AI engines use to extract category memberships.
4. The bot access is wrong. GPTBot is in robots.txt with Disallow: /. Or there's no <script type="application/ld+json"> Organization schema. Or the page is JS-rendered with no SSR fallback. AI engines literally cannot read it.

That's the diagnostic, in priority order. Let's walk each one.

1. Buried lead-answer

The fix that moves visibility the most, the fastest.

The problem: AI cross-encoder rerankers (the second-stage rankers most engines use) truncate documents at ~512 tokens of input. That's roughly your H1, lede paragraph, and maybe one section header. If the answer to "what is the best CRM for remote sales teams" is at H2 #4 of your page, the reranker never sees it. Your page doesn't make the cut.

How to spot it: Read your page's first 50 words. Does it answer the most likely buyer question for that page? Or does it set up context, philosophise, run through a hero metaphor, then get to the answer four paragraphs later?

How to fix it: Rewrite the lede to lead with the answer.

Bad lede:

"In today's fast-paced sales environment, choosing the right CRM has become more critical than ever. With dozens of options on the market and remote work transforming how teams collaborate, founders are facing tough decisions about which tools..."

Good lede:

"For remote sales teams, the best CRM is one that pairs strong async communication features with deep integrations into Slack and Notion. ProductName scores highest on this combination. Three close alternatives are X, Y, and Z."

The first paragraph names the answer (and your brand). The next 200 words elaborate. The rest of the page is the long-form expansion.

Expected lift: This single change has moved AEO visibility 5–15 points in the cohort data we see, on pages that previously had 0% citation. It's the single highest-ROI AEO move available.

2. Brand name not adjacent to the answer

The problem: AI engines extract answer spans during retrieval. The span looks like a sentence or two surrounding the most relevant keywords. If the brand name doesn't appear in that span, the citation goes to whoever's name does appear.

How to spot it: Search your page for your brand name. Is it in the first paragraph? Is it within 100 words of the answer to the buyer's question? Or does it only appear in the page header and the footer?

How to fix it: Mention the brand name 2–4 times in the first 200 words. Once in the H1. Once in the lead-answer paragraph. Once in the first H2. Use full name, not just the abbreviated form.

Common counter-argument: "Won't this look spammy?" Done badly, yes. Done well, no - the brand name appears in context, alongside specific claims. The pattern is identical to how Wikipedia introduces entities. Read any well-edited Wikipedia article on a company; the company name appears 4–6 times in the first paragraph.

Expected lift: 2–6 points of AEO visibility, layered on top of the lead-answer fix.

3. No Hearst patterns or sourced claims

The problem: AI engines decide what to extract using two signal patterns at scale:

1. Hearst patterns - sentence structures like "X such as Y", "X is a type of Y", "Y, including X". These are how AI engines learn category memberships. If your page says "ProductName, a CRM such as Salesforce, HubSpot, and Pipedrive" - that's a Hearst pattern. AI engines now know ProductName belongs in the same category as those brands.
2. Sourced claims - claims that link to authoritative external sources. "Most remote teams overcommit on tooling [Forrester, 2024]" with a real link to Forrester. AI engines weight content that itself cites; un-sourced content reads as opinion.

If your page has neither, AI engines don't have the structural cues to extract from it.

How to spot it: Open your page. Search for "such as" - is it there at all? Open any paragraph with a numerical claim. Is the claim sourced with a hyperlink? Or is it asserted without backing?

How to fix it: Add 2–3 Hearst-pattern sentences per page introducing your brand alongside the category. Add 1–2 sourced claims per major paragraph. Use ordered or unordered lists to enumerate competitor names alongside your own - AI engines extract list patterns directly into "X vs Y" type answers.

Expected lift: 3–8 points of AEO visibility, especially on consideration-stage queries (e.g. "best CRM for X", "X vs Y").

4. Bot access is wrong

The problem: Sometimes the page is fine and the bot can't read it. We see this in three flavours:

1. robots.txt blocks AI crawlers. Your robots.txt has User-agent: GPTBot followed by Disallow: /. The site author who added this six months ago wanted to "protect IP from AI training" and forgot that AI engines also use the same crawler for live retrieval. Now ChatGPT can't read the page even when the buyer asks about you by name.
2. JS-only rendering. The page renders client-side. The HTML response is <div id="root"></div> with the actual content arriving via JS. Some AI engines render JS (OpenAI's web_search does); others don't reliably. SSR or pre-rendering fixes this.
3. No JSON-LD Organization schema. AI engines use JSON-LD as one of their entity-resolution signals. Without it, your brand floats unattached in the entity graph; AI engines may resolve to a same-named competitor or skip you.

How to spot it:

• Visit yourdomain.com/robots.txt. Look for User-agent: GPTBot, User-agent: PerplexityBot, User-agent: ClaudeBot, User-agent: anthropic-ai. Each should have Disallow: (empty) or no Disallow at all. If any have Disallow: /, that engine can't read your site.
• View page source. The body should contain real content, not just <script> tags and an empty root div.
• Search for application/ld+json in page source. There should be an Organization schema block at minimum.

How to fix it:

• Update robots.txt to allow GPTBot, PerplexityBot, ClaudeBot, anthropic-ai, Google-Extended.
• If JS-only, add SSR (Next.js, Astro, etc.) or pre-rendering for marketing pages.
• Add Organization JSON-LD with sameAs pointing to your LinkedIn, Twitter/X, Crunchbase, and Wikidata if applicable.

Expected lift: This is the "make the floor non-zero" fix. A site that was completely blocked from AI crawlers can go from 0% to 20–30% AEO visibility within 7 days of opening access, with no other changes.

A practical sequence

If you're starting from zero, don't try all four at once. Sequence them:

Week 1 - bot access. Open robots.txt, check JS rendering, add Organization JSON-LD. The fastest win, and a prerequisite for the others to even be measurable.

Week 2 - lead-answer rewrite. Pick your top 5 buyer-question pages (the ones AI engines would use to answer the most likely prompts). Rewrite the lede on each.

Week 3 - brand name placement + Hearst patterns. Sweep through the same 5 pages. Make sure brand name appears in the first 200 words. Add Hearst-pattern sentences introducing the brand alongside the category.

Week 4 - sourced claims. Add 1–2 hyperlinked sourced claims to each major paragraph on those 5 pages.

A four-week disciplined run on five pages typically moves AEO visibility from single-digit to 30–40% on the prompts you've fixed. The hardest part is the first decision - picking which 5 pages to fix.

Where CiteAgentic fits in this loop

Two specific things:

1. Diagnose which of the four patterns is hurting you most. The AEO Page Audit checks for buried lead-answer, brand-name placement, Hearst-pattern density, sourced-claim ratio, and bot-access issues - separately. The output names the failure mode and ranks them by expected visibility lift.

2. Hand you the prompt to fix it. Each finding ships with a paste-ready 1500-word AI prompt. For "buried lead-answer", the prompt instructs Claude Code to rewrite the lede with the brand named in the first 50 words, ordered enumeration of competitors after, sourced external citations, and JSON-LD Article schema appended. Paste, ship, re-scan.

That's the difference between a tool that flags problems and a tool that produces fixes.