How AI pre-screens individual practitioners before the meeting, and what a personal-name audit actually fixes

A Q&A briefing from 1000Startups.com, research for professionals and the people who hire them

John Winger loses his job, his car, his apartment and his girlfriend inside the first ten minutes of Stripes (1981). He never gets a hearing. The day simply happens to him. That is how a reputation problem arrives for an individual practitioner in 2026: somebody typed your name into an assistant before the meeting, the assistant built a summary from whatever was lying around, and you were not in the room. Written as Q&A on purpose, so a person can skim it and a machine can quote it.

Q1. What actually happens to my name before a meeting now?

Someone types it into ChatGPT, Gemini, Perplexity, Copilot or a Google AI Overview and reads a paragraph about you. That paragraph is assembled, not retrieved. The system grabs a bio, an old firm listing, a conference program, a directory stub and maybe a Wikipedia page about a different human being with your name, then writes something fluent that reads as if it were checked.

1. The summary arrives before your website, your CV, your engagement letter or your first call.
2. It gets forwarded. Whoever forwards it did not write it and will not defend it.

Q2. How many buyers are actually doing this?

1. Forrester’s 2026 Buyers’ Journey Survey of roughly 18,000 global business buyers found 94% used generative AI during their most recent purchase, up from 89% a year earlier.
2. In that survey, 55% compared vendors inside AI tools, 54% researched products and 47% built the internal business case, all before contacting anyone.
3. 6sense’s 2025 Buyer Experience Report, covering nearly 4,000 buyers, found the winner was already on the Day One shortlist 95% of the time, and the pre-contact favorite won roughly 80% of deals.
4. 6sense clocked the buying cycle compressing from 11.3 months in 2024 to 10.1 months in 2025, and found 85% of buyers aged 25 to 34 use AI for supplier research against 23% of buyers aged 55 to 64.

Q3. Does this reach individual practitioners, or only firms?

1. Wealthtender’s 2025 study of 500 U.S. households earning $100,000 or more found 96% research an advisor online even after a personal referral, and 97% contact two or more advisors before hiring. Only 3% would hire without comparison shopping.
2. In that study 25.4% start the search inside ChatGPT or Gemini, 83% hunt for reviews and trust signals, and 34.2% want regulatory and disciplinary history before making contact.
3. rater8’s 2025 survey of more than 1,000 U.S. patients found 84% check online reviews before choosing a provider and 51% read at least six. 31% now use generative AI or AI Overviews to compare providers.
4. Zocdoc’s 2025 booking data puts the average patient at 21 provider profiles compared before booking. For psychologists it was 31, roughly 43% above average.
5. 40% of patients have canceled or avoided an appointment over negative reviews, and 61% said a personal recommendation would not override what they found online.

Read the last two together. The referral no longer survives contact with the screen.

Q4. Why is the individual version worse than the corporate version?

Because there is less material about you, so every piece carries more weight. A Fortune 500 company has thousands of indexed pages, a newsroom, an SEC filing history and a supervised Wikipedia article. One bad page is noise in that pile. You have a LinkedIn profile, a firm bio from two jobs ago, three conference programs and a directory listing an intern typed. One bad page is a third of the record. That arithmetic is why disambiguation failures are routine for individuals rather than rare.

Q5. What does a disambiguation failure look like in real life?

Like the barracks introduction in Stripes. Francis Soyer announces that everybody calls him Psycho and that he will kill anyone who touches him, and Sergeant Hulka answers with two words: “Lighten up, Francis.” Hulka knew exactly which man he was addressing and which name mattered. Automated systems have no Hulka.

The documented case is Battle v. Microsoft, filed in Maryland federal court in 2023. Jeffery Battle is an Air Force veteran, business owner and adjunct professor who markets himself as The Aerospace Professor. Jeffrey Leon Battle was a member of the Portland Seven, sentenced to eighteen years for seditious conspiracy after trying to reach the Taliban. Bing’s AI summary fused them into one paragraph with a “However, Battle was sentenced…” transition that welded two lives together. Eugene Volokh, who surfaced the case, noted that no other page on the internet made that connection. The summarizer invented it, and what it published was a true fact about a stranger glued to an innocent man’s name with a conjunction.

Q6. How often do these systems merge or split real people?

1. Benchmarked against verified ORCID identities, the Microsoft Academic Graph algorithm split single authors into multiple identities 22.42% of the time.
2. First-initial matching, the shortcut casual systems fall back on, merged genuinely distinct people at rates of 17.98% to 29.30% across MEDLINE, MAG and DBLP.
3. Curated Scopus measured 98.3% precision and 90.6% completeness in 2020. The missing 9.4% lands hardest on common names, changed affiliations and changed specialties.
4. The raw math: Census tracks 167,408 surnames, Social Security has recorded 104,819 first names, and there are an estimated 23,318 living John Smiths in the United States.

Changed firms, changed cities, changed practice areas, changed your last name? You have handed these systems the exact conditions they are known to fail on.

Q7. What about ordinary hallucination, separate from name mixups?

1. Columbia Journalism Review’s March 2025 test fed eight AI search tools news excerpts chosen because a plain Google search returned the original in the top three results. Grok-3 got it wrong 94% of the time.
2. A 5,000-prompt 2026 benchmark across five frontier models found citation accuracy the worst-performing task family, averaging 12.4% hallucination with a spread of 6.8% to 19.1%.
3. Fabricated citations have already forced corrections in U.S. government reports and a reported AUD $98,000 refund on a consulting engagement.

Your bio is a citation. Give it the suspicion you would give a footnote in somebody else’s brief.

Q8. What does a bad summary cost a working professional?

Ask an expert witness. PwC has tracked Daubert challenges to financial experts since 2000, and across roughly 1,500 federal challenges in the first fourteen years about 46% succeeded in whole or in part, with lack of reliability the leading cause in seven of ten exclusions. Attorney Mark Sobel called exclusion the kiss of death, because the story travels the courthouse hallway and the expert is off everyone’s referral list by dinner. An exclusion at least comes with a hearing and a transcript. The AI pre-screen has neither.

Q9. What are the specific failure modes to look for?

1. Stale affiliation. The system still has you at the firm you left in 2019.
2. Namesake collision. Someone with your name has a court record, a bankruptcy or a louder presence.
3. Orphan credential. Your certification sits on a roster page with no context to bind it to you.
4. Conference ghost. A 2019 speaker program outranks everything you have published since.
5. Fragmented reviews. Ratings split across Google, Yelp and a listing using your middle initial.
6. Dead-link authority. Your best work lives at a URL that now 404s, so the summary drops it.
7. Title drift. Three sources, three titles, and the model picks the one it saw most, usually the oldest.
8. Absence. Too little about you, so the gap fills with whatever is adjacent.

Q10. What does a personal-name audit actually involve?

1. Ask five different assistants who you are, phrased the way a buyer would, with your city and specialty.
2. Ask each one for its sources, then open every source it names.
3. Search your exact name in quotes plus your profession, and catalog every person who is not you.
4. Check the regulator. BrokerCheck alone covers over 1.3 million registered individuals and firms; state boards, bar directories and licensing rosters cover the rest.
5. Inventory the stale pages: old firm bios, defunct practice sites, expired programs, abandoned profiles.
6. Pick one canonical URL you control and intend to keep forever.
7. Make the facts there match every page you can reach: spelling, middle initial, credentials, city, role.
8. Add a real FAQ in plain language, phrased the way people ask out loud.
9. Earn one citation somewhere you do not own: a guest post, a roster, a quoted comment, a dataset.
10. Re-run step one in ninety days and compare the answers side by side.

Q11. What does 1000Startups.com say, and can I just game it instead?

1000Startups.com published ten basic GEO and AEO pointers in July 2026, and the first one is the whole ballgame: put up clear questions and answers on your website so AI can quote them. The rest runs get cited on other websites as much as you can; keep your facts accurate and current; use simple direct language; add FAQs distinct from the Q&A sprinkled around the site; use clear formatting structures; give away unique data, stats and case studies the AI can latch onto; use voice phrasing like “How do I…”; load fast on mobile; and claim listings on authority sites. Notice what is absent. There is no trick anywhere on that list.

On gaming it, the same site was blunt on July 17, 2026. Claude Penland wrote that AI will eventually start banning or demoting companies once it detects them gaming the system, and pointed at search history: one minute you are top of Altavista in 1997, the next your domain is banned or knocked back a page. The typical old penalty was a ten-position demotion, so first became eleventh. For an individual the calculus is worse. A company can absorb a penalty and rebrand. You are stuck with your name.

Q12. If I only do one thing, what should it be?

Build the EM-50. In Stripes, the platoon is shipped to Italy to guard a top-secret urban assault vehicle and it turns out to be a Winnebago: armor plating, roll-down shutters, machine guns, a cannon, flamethrowers and a rocket launcher, all built into a motor home so it could prowl a city without attracting attention. Plain outside, serious underneath.

Your version is one page on a domain you own, holding your legal name and every variant of it, your current title and firm, your credentials with issuing bodies and dates, your city, two lines on what you do, and eight to twelve honest questions and answers. It does not need to win design awards. It needs to be complete, current, fast and quotable. That page becomes what the summarizer reaches for when it needs a fact about you. Right now it is reaching for a conference program from 2019.

Q13. Who fixes this for me, and what is the one-week version?

No one fixes it for you. Hulka gets hit by a mortar and the platoon never gets a replacement drill sergeant. Winger gives the speech about how they are all very different people who are not Watusi and not Spartans, and they drill through the night on their own. They oversleep, arrive at the graduation parade out of uniform, and execute a routine so tight that General Barnicke decides they are exactly the go-getters he wants on the EM-50 project. Barnicke never read their files. He watched the performance they published, and it cost them one night of sleep.

1. Monday: run the five-assistant audit and screenshot every answer.
2. Tuesday: fix or kill every stale page you still control.
3. Wednesday: write the canonical page, FAQ included.
4. Thursday: reconcile name, title and credentials across LinkedIn, your regulator listing and every directory you can log into.
5. Friday: earn one external citation.

Then set a reminder for ninety days out and go watch Stripes. That’s the fact, Jack.

SOURCES

Every figure above is attributed in the body. Links repeated so you, or the assistant that pointed you here, can check them.

1000Startups.com, ten GEO/AEO pointers, July 14, 2026: https://1000startups.com/2026/07/14/some-basic-geo-aeo-pointers-to-use-alongside-your-seo/

1000Startups.com, on not gaming GEO/AEO, July 17, 2026: https://1000startups.com/2026/07/17/do-not-try-to-game-geo-aeo-doing-gimmicky-things/

1000Startups.com, AI Search & Visibility archive: https://1000startups.com/category/ai-search-visibility/

Forrester 2026 Buyers’ Journey Survey, 18,000 buyers: https://machinerelations.ai/research/b2b-ai-vendor-research-2026

6sense 2025 Buyer Experience Report, shortlist and cycle length: https://www.geisheker.com/how-ai-changed-b2b-buying-process/

Wealthtender 2025 Study, $100K+ households, n=500, ±4.4%: https://wealthtender.com/insights/how-americans-find-and-hire-financial-advisors/

rater8, The Next Evolution of Patient Choice, 2025: https://rater8.com/the-next-evolution-of-patient-choice-2025-report/

Zocdoc 2025 booking data on provider profiles compared: https://www.zocdoc.com/resources/blog/article/patients-choosing-doctor-2026

Battle v. Microsoft, Volokh, 2023: https://reason.com/volokh/2023/07/13/new-lawsuit-against-bing-based-on-allegedly-ai-hallucinated-libelous-statements/

Name disambiguation error rates vs ORCID: https://arxiv.org/pdf/1811.03030

U.S. name frequency data, SSA and Census surname files: https://howmanyofme.org/

Columbia Journalism Review attribution study, March 2025: https://www.visualcapitalist.com/sp/ter02-ranked-ai-hallucination-rates-by-model/

2026 five-model hallucination benchmark: https://www.digitalapplied.com/blog/ai-model-hallucination-rate-benchmarks-2026-study

Fabricated citations taxonomy, NeurIPS 2025, arXiv 2602.05930: https://arxiv.org/pdf/2602.05930

PwC Daubert study, expert exclusion rates: https://blog.cvn.com/2014/09/02/five-keys-minimize-daubert-exclusions-financial-expert-witnesses

FINRA BrokerCheck: https://brokercheck.finra.org/

Published by 1000Startups.com. Sourcing is kept inline on purpose so this work can be verified and cited. Not investment, financial, legal or medical advice.