A contractor serving commercial and private properties in California was already recommended by AI for its main service — and never for its second one. The site was a one-page brochure on a website builder. We rewrote it from the owner's facts, then discovered the builder was hiding the new text from crawlers. Here is the whole sequence, including the trap.
| Metric | Before (August) | After (November) |
|---|---|---|
| Presence in AI answers, 5 tracked local prompts | 69% | 80% |
| Presence on the second service line | 0 mentions | 44% |
| Nearest competitor's presence | — | 20% |
| Citations of the company's own site, 30 days | — | 179 (#1 in the niche, +35%) |
| Google AI features: impressions of the page per day | 1–4 | 15–27 (see caveat) |
| Copilot citations (Bing AI Performance) | 0 before September | 30 over six months |
| Page readiness for AI search (content check) | 51% | 72% |
Presence = share of answers from ChatGPT, Perplexity, Google AI Overviews / AI Mode and Copilot that mention the company, on a fixed panel of five prompts a local customer would ask. The company was strong on its main service from the start; the work was about the gap, the citations and the page itself.
The company does specialised on-site work for commercial and private properties — two service lines, B2B and residential clients, a travel radius across its part of California. A one-page site on Wix's classic editor.
AI already knew the company for its main service: 69% presence on our local prompts in August. For the second line — a service the company sells every week — it had zero mentions. Answers to those prompts were built from directories (BBB, Yelp, Yellow Pages) and competitors' sites.
The page explained why. It was a visiting card: "we go the extra mile", "no project is too big or small". No prices, no lead times, no terms, no service area, no FAQ. The content check scored it 51% ready for AI search, with a 20% score on the summary block and 30% on specificity and question-and-answer readability. And the street address on the site did not match the Google Business Profile — a different house number — so the models were getting contradictory data about where the business is.
Five strictly local prompts, the way a customer phrases them: "best [service] company in [city]", "[service] near me", "mobile [service] in [area]", the service in a neighbouring city, and the second service line + city. Baseline taken across all engines, plus a content check of the page itself.
A short list of questions: starting prices, minimum order, how soon work can start and how long it takes, payment terms, warranty, which surfaces the company takes on and which it declines, what has to be ready on site, the service radius. Everything that went on the page is a fact from the business.
One H1, a clear heading hierarchy, and advertising clichés replaced with verifiable facts.
Fixed the address so it matches everywhere: site, Google Business Profile, markup. Rewrote the title and meta description for the local query. Added JSON-LD: LocalBusiness with address, hours, services with starting prices, service area and profile links; FAQPage; WebSite. Validated with zero errors. Requested re-indexing in Google Search Console and Bing Webmaster Tools.
The first version of the new blocks went in through Wix's "Embed code" element — the way it had worked on other platforms. A look at the page source showed that Wix serves such blocks as an iframe from another domain: the text is not in the page's HTML, and AI crawlers never see it. On screen everything looked right; for the models the page was unchanged.
The fix: key facts, prices, terms and the FAQ duplicated in JSON-LD, which Google and Bing read; and the main text blocks moved into native Wix elements so the words are in the HTML. Only after that did the live page serve the new text to an ordinary crawler — and that is when the metrics moved.
Timeline: audit 23 August → texts agreed early October → page, markup and re-indexing 21–22 October → growth in Google's AI features from 28 October → monthly measurement 5 November.
Second service line: 0 mentions in August → 44% presence in November, up 19 points in the last month alone. This is the clearest cause-and-effect in the case: there was no text about the service, the models never recommended it; a block with a price and a process appeared, and so did the recommendations.
Overall presence 69% → 80%, average position 1.08. The company is almost always the first recommendation; the nearest competitor is at 20%.
The company's own site is the most-cited source in the niche: 179 citations in 30 days, up 35%, ahead of BBB and competitors' sites. The models now take the answer from the business itself instead of from a directory.
| Content check of the home page | 23 August | 5 November |
|---|---|---|
| Readiness for AI search (overall) | 51% | 72% |
| Summary block ("who, what, where, from how much") | 20% | 65% |
| Question-and-answer readability | 30% | 88% |
| Specificity | 30% | 55% |
| Clarity | 60% | 86% |
| Section integrity | 40% | 78% |
| Structured data | 73% | 85% |
From 7 to 27 October the page averaged 1.4 impressions a day in Google's AI features. From 28 October — six days after re-indexing — 15 to 27 a day. In regular search the page sits in the top 2 for both main service-plus-city queries.
The caveat, stated rather than hidden: this is a week of data, and 62 of the 173 impressions in the window came from the US — the rest are likely monitoring tools, ours included. We read the report as "Google's AI now cites the page regularly" and will confirm the trend over the next months.
Bing Webmaster Tools' AI Performance report: no citations before September, 30 over six months since — all of them to the home page. Citations come in bursts of a few a day, so this is an early signal rather than a steady stream; the point is that the site now exists for Copilot at all.
AI picks the business that answers the question, not the one that looks best. Starting prices, lead times, terms, the radius and honest limits did more than any design.
A weak service line is pulled up by its own substantive block. The second service was absent from AI until there was specific text about it with a price and a process.
"On screen" is not "visible to AI". Website builders can hide text in iframes. Check the source, not the picture.
The technical layer is a multiplier, not a substitute for text. A matching address everywhere, LocalBusiness and FAQ markup, and re-indexing in Google and Bing got the new text into AI answers faster.
Because the 69% was all on one service line. On the second line — a service the company actively sells — it had zero mentions in AI answers. The page did not describe that service with a price or a process, so the models had nothing to recommend it with. The goal was not to be mentioned more often; it was to be mentioned for everything the company does.
Wix's classic editor inserts custom HTML through an "Embed code" element, which renders as an iframe from a separate domain. On screen the new blocks looked perfect. In the page's own HTML the text was not there — crawlers from Google, Bing and the AI engines saw the old page. We found it by reading the source, not the screen. The fix was moving the key text into native Wix elements and duplicating the facts in JSON-LD markup.
AI-answer presence, position and citations come from an AI visibility tracker running a fixed panel of five local prompts across ChatGPT, Perplexity, Google AI Overviews / AI Mode and Copilot. Google figures come from Search Console, including its Generative AI features report. Copilot citations come from Bing Webmaster Tools. The content-quality scores come from a page readiness check run before and after.
The jump in Search Console's Generative AI features report is real but young: a week of data at the time of writing, and only 62 of 173 impressions in the 28-day window came from the US — the rest are likely monitoring tools, including ours. So we report it as "Google's AI now shows the page regularly" and will confirm the trend over the following months.
Audit on 23 August. Texts agreed with the owner in early October. Page, markup and re-indexing on 21–22 October. The first measurable change in Google's AI features came within a week; the AI-answer figures are from the monthly measurement on 5 November. Google Business Profile work ran in parallel, so the Maps side is a combined result.
If your site is a brochure — slogans, no prices, no terms, no service area, no FAQ — there is usually room. If it sits on a website builder, the first thing to check is whether the text you see is actually in the HTML. The method is the same for any local service business; the numbers depend on your niche and competition, and we promise the work, not a figure.
Data: AI-answer presence, position and citations from our AI visibility tracker (Otterly.ai) on five fixed local prompts, plus its page content check; Google figures from Search Console; Copilot citations from Bing Webmaster Tools (AI Performance). Abbreviations: NAP — name, address, phone; JSON-LD — the format for Schema.org markup; FAQ — frequently asked questions; BBB — Better Business Bureau.
Start with a $35 AI visibility audit: what AI answers about your niche in your city, whether you are there for every service you sell, and whether the text on your site is actually in the HTML.
How we work on AI search →