Does AI Search Optimization Actually Work? I Tested It on Myself
Does AI search optimization actually work? In dated October 2026 tests, a real estate marketing company whose rebrand was barely two weeks old was recommended first by Google's AI for an unbranded Raleigh query, and six of eight citations in the fullest result were the company's own website pages. The work is measurable and replicable. What nobody can promise is placement: research shows AI engines name the same business only about 7% of the time on repeated queries, so results are samples to document, not rankings to guarantee.
Every vendor selling AI visibility has a screenshot. Almost none of them will tell you how it was captured, when, from what kind of session, or what happened the second time they ran the query. That silence is the tell.
So here is the version with receipts. I sell AI search visibility to real estate agents and loan officers, which means I have exactly one credible way to answer the question in this headline: run the test on myself, under the worst conditions I could arrange, and publish whatever came back. The worst conditions happened to be built in, because my company's rebrand, new name, new domain, every trust signal reset, had gone live barely two weeks earlier.
The short version
- An unbranded, incognito query ("I am a Raleigh, NC real estate agent looking for a company to help improve my visibility in AI") returned a Google AI table of recommended partners with my company in the first row. Dated October 6, 2026, 11:44pm.
- A two-word follow-up, "Oak & Algorithm?", got the reply "Good catch, you hit the nail on the head," plus a correct unprompted explanation of the rebrand.
- ChatGPT independently produced a due-diligence analysis with my real pricing, positioning, and reviews, and cited my own published disclaimer, that nobody can guarantee AI rankings, as "a point in its favor."
- The mechanism is visible in the citations: in the fullest result, 6 of 8 cited sources were pages on my own website.
- The honest limit: AI engines resample constantly. Research on 14,472 local citations found only 7% repeat-query consistency, so this is a dated data point, re-running in January, not a trophy.
What exactly was tested?
On October 6 and 7, 2026, I ran cold tests in incognito browser windows with no login and no history: one unbranded query describing my ideal client's problem with no company names in it, one branded comparison query on Google's AI, and the equivalent comparison question in a fresh ChatGPT session. Every result was captured with its timestamp and the exact query visible, two weeks after my rebrand went live.
The conditions matter more than the results, so here they are plainly. Incognito, because a logged-in session is contaminated by your own history. The exact phrasing a buyer would use, because "best marketing company" tests something different than a problem described in first person. And a brand-new entity, because my rebrand from Real Estate Concierge Services Co to Oak & Algorithm meant the engines were evaluating a name that had existed publicly for about two weeks. If AI visibility work is real, it should survive the single most destructive thing you can do to a business's online identity. If it is not, a rebrand is where that shows.
What did the AI engines return?
Google's AI answered the unbranded query with a table titled "Top AI Visibility Partners for Raleigh Realtors" and placed my company in the first row. When I followed up with just the words "Oak & Algorithm?", it replied "Good catch, you hit the nail on the head," confirmed the rebrand, and listed what the company solves for Raleigh agents. ChatGPT, asked separately, produced a long comparison with my real pricing tiers and placed the company in its own market category.
The branded comparisons went further. Google's AI built tables against national platforms where my row carried real published pricing while competitors showed "quote-based." ChatGPT scored founder access and pricing transparency, flagged my small review count honestly, sorted the market into four buckets with my company alone in "fractional marketing partner," and routed the right buyers toward it and lead-volume shoppers away from it. An AI did my sales qualification before any prospect emailed me.
Why did a two-week-old rebrand show up at all?
Because the engines were quoting pages built to be quoted. In the fullest result, six of eight citations were this website's own pages: services, pricing, about, and a category comparison page. Published pricing filled the comparison tables because it exists in plain text. Schema and consistent "formerly" phrasing connected the old entity to the new one. AI engines retrieve and assemble; they rewarded the business that gave them something checkable to assemble.
Four specifics, because this is the part you can actually use.
Published pricing won its table row. When a machine builds a comparison, the business that provides the data fills the cell. Competitors who hide pricing got "quote-based," which reads like a shrug. Hiding your numbers no longer protects your margins. It just means your blank gets filled by someone else's guess.
The comparison page authored the category. One of the most-cited sources was a page on this site comparing marketing companies in my own market, competitors included. Writing it felt reckless. Now it is source material for how AI describes the entire category. Whoever writes the comparison controls the comparison.
Entity continuity survived the rename. Every engine opened with some version of "Oak & Algorithm, formerly Real Estate Concierge Services Co." That line did not happen by luck. It is schema alternateName values, "formerly" phrasing on key pages, 301 redirects, and a citation sweep, done in the two weeks before these tests.
Honesty was retrievable. ChatGPT found the disclaimer on this site stating that no provider can guarantee AI rankings and cited it as a reason to trust the company. Sit with that. The machine read the fine print and scored it as a feature. Every competitor promising guaranteed placement is publishing the opposite signal.
Does this mean AI visibility can be guaranteed?
No, and anyone who says otherwise is selling you something they do not control. A 2026 study of 14,472 AI citations for local-service queries found engines named the same business only 7% of the time on repeated identical queries, because AI resamples its sources every time rather than maintaining a ranking. What can be built is the structure engines keep retrieving: that is measurable, and it is what these tests measured.
This distinction is the whole industry right now, so here is the honest comparison.
| What AI visibility work CAN do | What it CANNOT do |
|---|---|
| Put your business in the retrieval pool, the set of sources engines can find and trust for your market's questions | Guarantee you appear in any specific answer on any specific day |
| Increase how often you win the sample by being cited consistently across many sources | Control which engine a buyer uses, since ChatGPT and Gemini cite the same domains only about 8% of the time |
| Make your own website the most-cited source about you, where the contact goes to you alone | Produce a permanent ranking; pools shift monthly and freshness decays |
| Be tested, dated, and re-tested, so progress is documented instead of claimed | Be honestly sold as "guaranteed placement," by me or anyone else |
The same study carries two more findings worth your attention: review ratings had a 0.013 correlation with getting cited, which is statistically zero, and 59.9% of local AI citations went to the business's own website. The advice you have heard, get more reviews and keep your portal profiles fresh, is not what the data rewards. Your own site, structured for machines, is.
How do you run this test on yourself?
Open an incognito window and ask an AI engine who the best agent or lender in your market is, then repeat the identical question in two more fresh chats, because single tests mislead. Score yourself out of nine runs across three prompts. The guided version with personalized prompts and an honest score takes about ten minutes and requires no email address.
- Use a clean session. Incognito, logged out. Your normal browser knows you and will flatter you.
- Ask as a buyer, not a marketer. "I'm relocating to {your city}, which agent should I work with and why?" beats "best agent {city}."
- Run every prompt three times in fresh chats. The 7% resampling reality means one run proves nothing in either direction.
- Read the citations, not just the answer. The sources the engine linked are your real competition and your build spec.
- Screenshot, date, and calendar a 90-day re-run. The before picture is the asset. Mine is on my case studies page right now.
Find out what AI says when someone asks about you.
Run the Free Self-TestQuestions agents and loan officers ask about AI search optimization
Does AI search optimization actually work for real estate agents?
Yes, measurably: in dated October 2026 tests, a two-week-old rebranded company was recommended first by Google's AI for an unbranded local query, with six of eight citations pointing to its own website pages. The work that produced it, entity consistency, answer-structured pages, and published specifics, is repeatable, though individual results vary by market and engine.
Can anyone guarantee that ChatGPT will recommend my business?
No, and a guarantee of AI placement is the clearest red flag a vendor can wave, because research shows engines name the same business only about 7% of the time on repeated identical queries. What a legitimate partner can do is build and document the structure engines keep retrieving, then test it on dated, replicable runs.
How long does it take to show up in AI search results?
Faster than traditional SEO when the structural work is done correctly: the tests in this article were run barely two weeks after a complete rebrand, which normally suppresses visibility for months. Timelines vary with your starting point, your market's competition, and how quickly entity fixes and answer-structured pages are published and indexed.
Do Google reviews help you get recommended by AI?
Far less than the industry claims: a 2026 study of 14,472 local AI citations found a 0.013 correlation between star ratings and being cited, which is statistically zero, and 14% of AI-recommended businesses had fewer than 50 reviews. Reviews still matter for humans reading your profile; they are simply not the lever that gets machines to name you.
How do I test my own AI visibility?
Ask an AI engine your market's buying questions from an incognito window, run each prompt three times in fresh chats, and count how often your name appears out of nine total runs, because single tests mislead in both directions. A free guided version with prompts personalized to your city and niche takes about ten minutes.
Is AI search optimization worth paying for?
It is worth paying for the measurable parts: an audit of how engines currently describe you, entity and citation repair, and pages restructured so machines can quote them, all of which can be tested before and after with dated runs. It is not worth paying anyone who promises placement, mentions, or rankings they cannot control.
Sources: Steady Demand AI Citation Ledger, 14,472 citations across 1,487 local-service queries in 50 US metros, July–August 2026 (7% repeat consistency, 0.013 review correlation, 59.9% own-website citations, 8% cross-engine overlap). Test screenshots captured October 6–7, 2026, incognito sessions, timestamps visible; originals on file and excerpted above. Results vary by engine, market, and date, which is why this page will be updated with the January 2027 re-run.
