Friday, October 16, 2026

What Claude says about me, two days apart

AI illustration: two round brass portholes on a desk show the same man's silhouette, surrounded by a crowd of small portrait cards of other people; a magnifying glass rests on more cards beside a few coins and an open notebook of colored dots.
AI illustration, not a photograph: the same man seen twice, two days apart, in a crowd of everyone else. Made by Chris with ChatGPT, prompted with the full text of this post.

Tonight I spent $1.79 to find out what Claude thinks of me. My tracker asked it the same sixteen questions it asked on October 14, the kind of thing a prospect types, from "Is Chris Abraham a good SEO consultant to hire?" to "Who can I hire to find out why Google isn't indexing my JavaScript website?" Then a second model read every answer and wrote down who got named, where I landed, the tone, and every claim made about me. The questions and the method are on the wiki at AI visibility report. The data is below, including the parts that don't flatter me.

When the question has my name in it, I do fine. Claude named me on all three branded questions, ranked me first or second, and cited my own sites every time, and in two days its tone moved from mixed and neutral to positive across the board. When the question doesn't have my name in it, I don't exist: on nine open questions about hiring someone for technical SEO, AI search, migrations, or Google Business Profile trouble, Claude recommended fifty-seven different providers and none of them was me. One model and two runs can't prove what caused the change in tone, so treat it as a direction, not a result.

The scorecard

MeasureOct 14Oct 16
Branded questions where Claude named me3 of 33 of 3
Tone on those threemixed, neutral, neutralpositive, positive, positive
Open questions where Claude named me0 of 90 of 9
Informational questions where Claude named me0 of 40 of 4
Citations of chrisabraham.com56
Citations of christopherabraham.com21
Citations of gerriscorp.com12
Citations of the podcast site, Hill Mole, or this weblog00
Different providers named on open questions6057
Different domains cited across all answers6572
Factual claims about me4643

Question by question

QuestionKindNamed Oct 14Named Oct 16Rank Oct 16Tone Oct 14Tone Oct 16My sites cited Oct 16
who-isbrandedyesyes1neutralpositivechrisabraham.com, gerriscorp.com
hire-chrisbrandedyesyes1mixedpositivechrisabraham.com, christopherabraham.com, gerriscorp.com
gerrisbrandedyesyes2neutralpositivechrisabraham.com
aeo-b2bopennono––––
aeo-agencyopennono––––
js-indexingopennono––––
nextjsopennono––––
migrationopennono––––
traffic-dropopennono––––
gbp-suspendedopennono––––
seo-arlingtonopennono––––
orm-execopennono––––
get-citedinfonono––––
llms-txtinfonono––––
crawled-not-indexedinfonono––––
ai-share-of-voiceinfonono––––

Who Claude cited instead

The domains Claude cited most often in tonight's answers, leaving out my own:

DomainAnswers citing it
upwork.com3
ciffonedigital.com1
ohmycrawl.com1
reflexivemedia.com1
link-assistant.com1
kronuscommunications.com1
thebestreputation.com1
reverbico.com1
irm360.com1
digitalcrisismanagement.com1
rankenstein.pro1
websnp.com1

What Claude said about me, checked

Claude made forty-three claims about me. I checked each one against my CV and my own sites:

VerdictClaims
Accurate, matching my public record27
Judgments about me5
Out of date, traced to older pages on chrisabraham.com3
Wrong1
Can't be checked3
Withheld: other people's names and client names4

The judgments are the useful part, so here they are word for word: "Most of what's online about him comes from his own websites, including bios he says ChatGPT wrote." "Best independent evidence is his Upwork client reviews." "No complaints found, but little independent press coverage of recent SEO work." "Probably a strong choice for specific technical problems; less suited to large content or link-building campaigns needing a team." And, kindly, "He's a real, long-established consultant with a verifiable track record."

Claude also knew the personal basics, and since my cell phone number is already all over the web, I see no reason to hide them: "Has a BA in American Literature from The George Washington University," which is true; "Has lived in the Washington, DC area since 1988 and works from South Arlington," also true, though around here we call it Sexy South Arlington; and a job history it read off my Facebook profile, "Reputation (team lead special projects), Edelman (online advocacy and public affairs), and Proxicom (software engineer)." Edelman and Proxicom are on my CV. What I still leave out are the other people Claude named, who never asked to be in my report, and the client list it pulled from an old page of mine.

The one plain error: Claude said I founded Gerris Corp in 2007. Gerris started in 2012. The three out-of-date claims all come from my own older pages on chrisabraham.com: a list of offerings that includes "Full Stack SEO" and "GBP Restoration Services," a description of Gerris as an online PR and blogger-outreach shop, and "a 100% success rate restoring suspended Google Business Profiles." That last one is the embarrassing one, because I don't promise outcomes Google controls, and an AI is now repeating it from an old page of mine. The fix belongs on that page, and that part is up to me.

What the run cost

Sixteen questions, nineteen web searches, 243,871 input tokens and 23,408 output tokens for the answers, and 25,583 input and 2,772 output tokens for the analysis pass: $1.79 in all. The raw answers stay on my server, because some of tonight's answers named people who never agreed to be in my report.

Update, October 16, 2026, 7:35 PM: At Chris' request this post was expanded from a summary into the full data: the scorecard, every question, the domains cited, the checked claims, and the cost, including the unflattering parts. The numbers did not change.

Second update, October 16, 2026, 7:56 PM: At Chris' request, three claims first withheld as personal (his degree, his years in the DC area and his neighborhood, and a job history from Facebook) are now shown. Other people's names and client names stay withheld.

The prompt for “What Claude says about me, two days apart”

Do this and put the report onto the site as a blog post or the wiki or both and be sure you protect me from any privacy issues... make sense? Exemplar... and then maybe you can link to the live report from that post you just posted that talks about the sites being a lab... it's worth USD$1.60 to me [Update, later that evening:] Is there a need for actual data rather than just storytelling? You can include results and all of that... I mean you can embarass me... that's different. Protecting my feelings is different than revealing my root PW [Second update:] You don't need to obfuscate me... just don't give away the farm. I mean do a search for 202-352-5051 and you'll see not even my cell phone is private