How to fact check a confident AI answer about SEO before you act on it
I asked an AI search assistant whether ending every page title with a pipe and the company name hurts a website. The answer came back polished, cited, and full of percentages. When I pushed my own opinion back at it, it declared that opinion “100% true” and invented psychology to support it. The recommendation itself was reasonable. A good share of the evidence was not. This is the routine I use to separate the two before anything reaches a client's site.
Turn the answer into a list of claims
Paste the answer into a document and give every sentence that asserts a fact its own line. In this exchange the list read: titles are cut off after about 60 characters; Google rewrites more than 75 percent of titles; Google strips pipes and brand names; Google displays the site name on its own; repeated suffixes cause duplication penalties and dilute relevance; searchers read a pipe as a sign of automated spam.
A sound recommendation can rest on claims that are half wrong. Grading each line separately stops a good conclusion from carrying bad evidence into your report.
Grade every claim by where it comes from
Each line goes into one of four bins: the platform's own documentation, an independent study with a published method, a vendor test, or nothing traceable. Here is how this exchange graded out.
| Claim | What I found |
|---|---|
| Titles are cut off at about 60 characters | Google's documentation sets no maximum length and says the displayed title is shortened to fit the device. Sixty characters is an industry habit, not a Google limit. |
| Google rewrites over 75 percent of titles | I could not trace that figure to a primary dataset about pipes or brand names. The best documented study I found, by Zyppy, measured 61.6 percent of 80,959 titles on 2,370 sites, using desktop results from early 2022. Titles longer than 70 characters were rewritten 99.9 percent of the time. |
| Google strips pipes | The same study found pipes removed or replaced in 41.0 percent of titles that used them, often swapped for a dash. Dashes were removed 19.7 percent of the time. |
| Google strips brand names | The study names missing or superfluous brand names as a rewrite trigger and gives no percentage. Google says it may omit a site name that repeats the one already shown. |
| Google shows the site name separately | Supported by Google's documentation. |
| Repeated suffixes cause penalties or crawling problems | Nothing traceable. Google warns that boilerplate titles make results hard to tell apart, which is about usefulness to searchers, not a penalty. |
| Searchers read a pipe as spam | Nothing traceable. I found no study measuring that reaction. |
Follow every number back to its population
A statistic describes the set of things someone counted, at a particular time, in a particular way. Ask who measured it, when, on which device, and what exactly was counted. A rewrite rate for all titles says nothing specific about pipes. A desktop sample from early 2022 may not describe mobile results today. When an assistant cannot show where a number came from, or the source measures something else, drop the number and keep the point in plain words.
Distrust agreement that grows when you push
The clearest warning in this exchange came after I shared my own view. The follow-up adopted my wording, raised its certainty, and added mechanisms nobody had measured. An assistant that grows more confident because you sounded confident is reflecting you back at yourself. Open a fresh conversation and ask the opposite question: what is the case for keeping the company name in every title? If the answer flips that easily, the evidence was never doing the work.
Settle it with your own data
For title changes, the deciding evidence sits in your own Search Console. Change one group of pages, leave a comparable group untouched, and compare clicks and click-through rate across several weeks. A controlled result from your own site outranks any general statistic, accurate or invented. My guide to diagnosing a Google traffic drop in Search Console covers lining changes up with dates, and the one on measuring AI citations, referrals, and leads applies the same discipline to AI search reporting.
What survived the check
After grading, the defensible advice fit in a few lines: write a specific title for each page, include the company name where it helps someone recognize or choose the result, keep titles concise, and measure the change. Dropping a repeated suffix remains a legitimate editorial call. It simply has to be presented as judgment backed by a test, never as escape from a penalty that does not exist.