This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
The most useful test question in the Washington Post’s recent review of Google’s AI search was also the most ordinary one: what is Mark Zuckerberg’s net worth?
Regular Google answered with a single figure. Google’s Search Generative Experience, the experimental AI version, answered with a string of hourly, weekly and daily numbers that did not add up, and, according to the review by Geoffrey Fowler, appeared to rely on a ZipRecruiter page about “Mark Zuckerberg jobs.”
Fowler’s broader finding, after nearly 11 months of public testing, was that SGE still sometimes makes up facts, misreads questions, serves outdated information and elevates low-quality sources. The Post cited an SE Ranking analysis of 100,000 keyword searches in which Quora was the site SGE linked to most often, with LinkedIn and Reddit also near the top. Google disputed the outside research, saying it was based on a narrower set of searches than Google sees, but did not share data of its own.
Fowler also noted that Google had started showing AI answers in the main results to some people who had not opted into the test. That is the part communications teams should pay attention to.
Why the net worth example matters
Most people will read that example as a funny glitch. I read it as a preview of where executive and corporate facts are most fragile.
Net worth, revenue, headcount, founding dates, ownership stakes, settlement amounts. These are numbers that already exist across many pages in slightly different forms. A generative system that blends sources can produce a confident figure that no single source actually states. For a public company or a well-known executive, a wrong number in a prominent answer is exactly the kind of detail a journalist, investor or recruiter ends up repeating.
What to do while SGE is still an experiment
The good news is that the practical work is mostly the same work that has always protected a reputation in search. The difference is that the margin for neglect is getting smaller.
Check your own names in SGE. Search the company, the CEO and the key products with the experiment turned on. Note what it says and which sources it links. This takes an afternoon and tells you more than any forecast about AI search.
Fix the facts at the source, with dates. If your own site describes your company inconsistently across pages, a blending system has inconsistency to work with. Put the important facts in one clear, dated place, and keep them current. As an earlier post noted last August, dates are becoming part of how these answers signal reliability.
Take forum and profile pages seriously. If Quora, Reddit and LinkedIn are among the sources SGE reaches for, then outdated or speculative posts on those platforms are no longer harmless background. An accurate LinkedIn page for an executive is a modest effort with a real payoff.
Strengthen third-party corroboration. Generative answers lean on consensus. A fact that appears in your press release and nowhere else is weaker than a fact confirmed by credible independent coverage. The corroboration problem has not gone away just because the answer now arrives first.
Do not chase “SGE optimization.” The product is still in Labs and has changed repeatedly. Tactics built around its current quirks will age quickly. Accurate, consistent, well-sourced information will not.
It would be easy to read the Post review as a simple verdict on Google’s AI. The more useful reading is that the system will summarize whatever the web gives it, including the weakest page available. The reputation job is making sure that page is not the one about you.