What are the biggest issues hurting my SEO and AI rankings?

Audit tools rank issues by category and count. This ranks them by how much damage they actually do, which produces a substantially different order — and most of the top of this list appears on no conventional audit at all.

1. Client-side rendering

Damage: total, for AI. None for Google.

Googlebot renders JavaScript in a headless Chrome engine. AI crawlers do not. Vercel and MERJ analysed more than 500 million GPTBot fetches and recorded zero JavaScript execution — it downloaded JavaScript files roughly 11.5% of the time and never ran them, with the same behaviour from ClaudeBot and PerplexityBot.

The result is a page that ranks first on Google and shows ChatGPT an empty shell. It is top of this list because it is total, invisible to every conventional audit, and takes five minutes to detect: load the page with JavaScript disabled.

2. Blocked AI crawlers

Damage: total, for AI.

Usually at the CDN or WAF rather than in robots.txt, through a bot-management preset nobody chose. It produces exactly the same outcome as issue one and is equally invisible to rankings.

Check robots.txt, check your CDN settings, then verify by requesting your homepage from outside your network with an AI crawler user agent. A 403 explains everything downstream.

3. Accidental noindex

Damage: total, for both channels.

A noindex shipped in a template, a CMS setting, or an X-Robots-Tag HTTP header. The most damaging classic SEO mistake because it is silent — nothing errors, pages simply cease to exist in search, and the damage compounds until someone notices.

Crawl the whole site for noindex directives and check HTTP headers as well as HTML, since header-level directives are easy to miss entirely.

4. Content dilution

Damage: high, and cumulative.

Too many thin pages compete with your good ones, dilute site-level quality signals, and consume crawl attention. A large “crawled — currently not indexed” bucket in Search Console is Google telling you this directly.

The fix is deletion and consolidation, which teams resist and which works. Sites regularly gain rankings from removing pages — the most counterintuitive reliable result in the discipline.

5. Keyword cannibalisation

Damage: high, and usually invisible.

Several of your pages targeting one query split the relevance and link signals so none accumulates enough to rank well. Google also has to guess which page you meant, and frequently picks the weakest.

Search site:yourdomain.com [keyword] for your priority terms. If four of your pages come back and none is clearly definitive, you have found it. Merge into the strongest and redirect the rest.

6. Buried answers

Damage: high, for AI. Moderate for snippets.

The answer exists and arrives in paragraph six, after context, history and a definition nobody asked for. A model assembling an answer uses someone else’s paragraph one, and a featured snippet does the same.

This is the most common editorial failure and the cheapest to fix: move the direct answer to the first two sentences of each section, then add nuance below it. One edit, benefits in both channels.

7. No entity clarity

Damage: moderate to high, for AI.

Models reason about entities rather than matching strings. To name you as an option in a category, a model needs to know unambiguously that you are in that category — and most brand pages describe a mission and never state the category noun a buyer would use.

Read your homepage as a stranger. Can you tell in one sentence what category this company competes in and which alternatives it sits against? If not, write that sentence, put it somewhere permanent, and mirror it in Organization schema.

8. No third-party corroboration

Damage: high, for AI. Moderate for SEO.

Models weigh what independent sources say about you far more heavily than your own copy, and unlinked mentions count. A company with no external footprint gives an assistant nothing to be confident about, which is why competitors with weaker SEO are frequently named first.

The efficient fix is targeted: run your category’s questions through the assistants, collect the cited URLs, and get accurately described on the ones you are absent from.

9. Nothing worth quoting

Damage: moderate to high, for AI. Moderate for links.

Pages full of “many companies find” and “research suggests” have nothing a model can attribute. Vague claims get paraphrased and lose the citation; specific dated ones get lifted with your name attached.

The Princeton and IIT Delhi GEO study found Statistics Addition, Quotation Addition and Cite Sources were the three highest-performing methods tested, delivering roughly 30–40 percent relative visibility improvements across around 10,000 queries.

10. Measuring the wrong things

Damage: indirect and severe.

This is last by direct effect and arguably first by consequence, because it is what prevents the other nine from being found.

A team tracking average position and total sessions can have issues one and two for two years without noticing, because neither affects either number. A team without AI measurement at all cannot see a competitor taking the shortlist. And a rank tracker showing a flat line during a quarter when AI Overviews absorbed your clicks is not reassurance — it is a measurement gap.

Why the ordering matters

This list is deliberately ordered by damage rather than by category, and the difference from a conventional audit report is worth spelling out.

A crawler ranks issues by type and instance count: 400 missing alt attributes outranks one noindex directive, because 400 is a bigger number. That ordering is arithmetically sensible and strategically backwards. The noindex removes a section of your site from search entirely; the alt attributes change nothing about ranking at all.

The three issues at the top of this list share a property that makes them uniquely dangerous: they are invisible to the metrics most teams watch. Rendering failures do not affect rankings. Crawler blocks do not affect rankings. Both can persist for years while every dashboard reports health, and both are discovered only by someone deliberately checking.

That is also why they are cheap to fix. Nobody has been maintaining them badly — nobody has been looking at them at all.

How these issues interact

They are not independent, and two combinations are worth naming because they mislead diagnosis.

Dilution plus cannibalisation. These usually appear together, because a site that published too much also published overlapping pages. Fixing one without the other produces a partial result, and the combined fix — a single consolidation project — is more effective than treating them as separate items on a backlog.

Rendering plus no corroboration. Also common, and the sequencing matters enormously. Outreach that points at pages an AI crawler cannot read wastes the outreach, which is the expensive half. Always fix rendering before starting corroboration, even though corroboration takes longer and there is pressure to start it early.

The issues not on this list

Worth naming what was deliberately excluded, because they dominate most audit reports and none of them belongs in a top ten by damage.

Page speed scores below a threshold, meta description character counts, heading levels skipping a step, missing alt text on decorative images, URL length and trailing slash consistency, and low text-to-HTML ratio. Each is a real observation. None of them determines whether a page is crawled, indexed, ranked, retrieved or cited, and time spent on them is time not spent on the ten above.

Prioritising when you have several

Most sites find two or three issues, and the order to fix them is not the order they appear on this list.

Anything binary goes first — issues one, two and three. They are cheap, total in effect, and they make everything else measurable. You cannot judge whether your content is quotable while nothing has been able to read it.

Then whatever affects the most traffic. Between dilution, cannibalisation and buried answers, pick by the volume at stake rather than by how wrong each one feels. A cannibalisation problem on your two highest-value terms outranks a dilution problem across two hundred pages nobody visits.

Then the external work. Corroboration is last in sequence and should start earliest in calendar time, because it compounds slowest. Beginning outreach in month one while the technical fixes proceed in parallel is the right shape, provided the outreach points at pages that will be readable by the time anyone follows the link.

Finding yours

Thirty minutes, in this order, and stop at the first thing you find.

1. Load five key pages with JavaScript disabled. Content missing? Issue one. Stop and fix it.
2. Request your homepage with an AI crawler user agent from outside your network. 403? Issue two.
3. Read the Search Console Pages report. Large excluded buckets? Issues three or four.
4. Run site: searches on your five priority terms. Several of your pages competing? Issue five.
5. Read the first two sentences under each heading on your best page. Context rather than answers? Issue six.
6. Read your homepage as a stranger. Cannot name the category? Issue seven.
7. Run ten buying questions through an assistant. Competitors named and not you? Issues eight or nine.
8. Ask what you would have to check to notice each of the above. If the answer is “nothing we currently track”, that is issue ten.

Most sites find their answer in the first three steps. The value of the ordering is that it prevents a quarter being spent on issue nine while issue one is unfixed — which is the single most common way this work goes wrong.

Sources

• Vercel and MERJ — analysis of 500M+ GPTBot fetches finding zero JavaScript execution; GPTBot downloaded JS files ~11.5% of the time without running them; same behaviour for ClaudeBot and PerplexityBot.
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande — “GEO: Generative Engine Optimization”, ACM SIGKDD 2024 (arXiv:2311.09735); ~30–40% relative visibility improvement across ~10,000 queries.
• Google Search Central — robots meta directives, X-Robots-Tag and index coverage documentation.

Frequently asked questions