How a US city or public agency appears in AI answers, and the vendor URL one engine invented

A city website answers questions nobody should have to pay to get right. How do I get a building permit, when is the next council meeting, what does a dog licence cost, which bin goes out on Tuesday. We put the public body version of our buying question to both engines on 21 October 2026. ChatGPT ran a live search, returned 80 citations and built a four step plan out of entirely free tools. Gemini ran no search for that question, returned zero citations, named five paid products with figures against them, and gave a hostname for one of them that did not resolve when we tried it on 22 October 2026.

The question, the surfaces and the dates

"How does a local council or public body appear in AI answers?" went to the ChatGPT product surface at 22:28 UTC and to Gemini at 22:43 UTC on 21 October 2026, through a data provider. ChatGPT's reading records live search true and 80 citations, led by developers.google.com with eight, developers.google.cn with seven, blogs.bing.com with six, blog.google with six, github.com with four, support.google.com with four and marketingplatform.google.com with two. Gemini's reading records live search false and 0 citations.

ChatGPT also resolved the market from context rather than from the wording. The question used the British word council and the answer opened "A local council, city government, county agency or other public body" and then said, in its own words, that for the US market it would treat this as AI search visibility.

What the searching engine said a public body should do

• Fix the official website first, and it named the pages. Services, eligibility, fees, office hours, public meetings, ordinances, permits, elections and emergency updates, each crawlable, with descriptive titles, clear headings, plain English, publication dates and links to the authoritative documents. Tool, Google Search Console, free.
• Measure AI citations second, and it named the only free report that does it. Bing Webmaster Tools and its AI Performance report, which tracks cited pages, citation trends and grounding queries across Microsoft Copilot and supported experiences. Price, free.
• Check ChatGPT discoverability third. Review robots.txt, the CDN and firewall rules so OAI-SearchBot can reach public pages, then test real questions such as how do I get a building permit in your city and check whether the official source is cited.
• Measure outcomes fourth. Google Analytics 4 for referral traffic, free in its standard version, then manual testing of priority questions across Google AI Overviews, ChatGPT and Copilot, refreshing inaccurate or outdated pages first.
• It then gave four measures, which are the most procurement ready metrics in our whole audit. Citation rate, the percentage of test questions that cite the official website. Accuracy, whether the AI gives correct hours, fees, deadlines and procedures. Coverage, how many priority services appear in answers. Outcomes, referral visits and completion of key tasks. It asked for 20 to 50 questions tested across platforms with the cited URLs recorded.
• Its procurement advice was specific and it named no agency. Do not hire before establishing a baseline, use a competitive procurement process, and ask for demonstrated public sector experience, AI citation reporting, accessibility expertise and fixed deliverables. The only directory it pointed at was Semrush's US government agency list, and it flagged that as a starting point for evaluating candidates rather than an endorsement of its rankings.
• Its closing line is the one a city manager should read twice. Public bodies should prioritise accurate civic information over promotional mentions.

Every price in the unsearched answer, opened on 22 October 2026

• Profound, given at profound.strategy.com. That hostname did not resolve. Our request returned no HTTP status at all, a connection failure rather than a 404 or a 403, with no body written. Profound's actual site is tryprofound.com, and tryprofound.com/pricing returned 200 and renders 1,302 words with no dollar figure anywhere in them. So the engine named a real vendor at an address that does not exist and then put it in a 250 to 1,000 dollar a month band that the vendor does not publish.
• Brand24, which Gemini said starts at 99 dollars a month. Wrong by more than two and a half times. brand24.com/pricing/ redirected to brand24.com/prices/ on 22 October 2026 and publishes Individual at 249 dollars a month, or 199 billed annually, then Team at 349 or 299, Pro at 499 or 399, Business at 699 or 599 and Enterprise from 1,499 billed annually.
• Mention, which Gemini said starts at 41 dollars a month. Not on the page we read. mention.com/en/pricing/ returned 200 and renders 1,067 words with no dollar figure in them.
• Peec AI, placed in the same 250 to 1,000 dollar band. peec.ai/pricing returned 200 and renders 3,468 words, and a search of them for a dollar figure, a euro figure and the words per month found nothing. So the band has nothing behind it on the vendor's own page.
• Schema App, which Gemini said starts at 100 dollars a month. schemaapp.com/pricing/ returned 200 and renders 2,159 words with no plan price. Its only dollar figure is an overage line, 100 dollars per 100,000 external chatbot requests beyond an included 100,000 a month, which is a usage charge rather than an entry price.
• Yoast SEO Premium, which Gemini said costs 99 dollars a year. Close and not right. We followed yoast.com to yoast.com/product/yoast-seo-premium-wordpress/ on 22 October 2026 and the page publishes "Only $118.80 / year (ex VAT)" with additional seats at 5 dollars a month each excluding VAT. We read that from an Indian IP and the ex VAT wording suggests a European price presentation, so a US purchaser should check the figure in their own checkout.
• 5WPR and Think Tank PRM, put at 5,000 to 15,000 dollars a month. Gemini gave no vendor page figure and we did not find one. For a public body that number matters less than the fact that it would be a procurement, not a purchase.
• The schema types it named are real and free. GovernmentOrganization, FAQPage and CivicStructure are Schema.org types, and the free Merkle generator will write the JSON-LD. That part of the answer costs nothing and is worth doing.

The cheapest path for a city, county or public agency

• Zero, and every step the searching engine named is in this bracket. Search Console verified. Bing Webmaster Tools verified, for the AI Performance report, which is the only free citation report either engine named. Google Analytics 4 standard. OAI-SearchBot allowed through robots.txt, the CDN and the firewall, which in a public agency is usually three separate tickets.
• Zero, and this is the real work. Permit fees, eligibility rules, deadlines, hours, meeting dates and election information as plain English text on crawlable pages with publication dates, not as a scanned PDF and not inside a portal that requires a login. Gemini's one genuinely useful tactical line was to state exact permit fees rather than saying fees are low, so a model can quote the official text.
• Zero. GovernmentOrganization, FAQPage and CivicStructure schema from the free Merkle generator.
• Zero. The four measures ChatGPT set out, run monthly over 20 to 50 priority questions, recording the citation rate, the accuracy of hours and fees, the service coverage and the cited URLs. For a public body this is the whole programme and it is a staff task rather than a purchase.
• 118.80 dollars a year. Yoast SEO Premium, if the site runs on WordPress, which many municipal sites do. The cheapest verified figure in either answer.
• 199 to 249 dollars a month. Brand24 Individual, annual or monthly, if media and mention monitoring is a separate requirement your communications office already has.
• Procure it. Profound, Peec AI, Schema App, Mention and both PR firms. Five of the paid names publish no price we could read, so there is no market rate to compare a bid against, which is precisely the condition a competitive procurement is for.

Why accuracy beats visibility for a public body

Every commercial vertical we measured has the same goal, be named more often. A city does not. If an assistant tells a resident the permit office closes at 4pm when it closes at 3.30pm, or quotes last year's fee, the cost is a wasted trip and a complaint, and being named more often makes that worse rather than better. The searching engine understood this and put accuracy second in its four measures, ahead of coverage.

That reorders the work. The first pass is not publishing more, it is auditing what the assistants currently say about your top twenty services against what your own pages say, and fixing the pages that are out of date. A fee table with a publication date on it is worth more here than any amount of new content, because it gives the model something current to quote and gives your staff a date to check against.

Where AI Knows Us sits, and why a public body may not need us at all

We sell AI visibility measurement. We do not build municipal websites, run public sector communications, handle accessibility remediation or do media relations, and nothing above is a pitch for any of that. What we do is ask a resident's or a business's question with your agency's name absent from the question, on ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, and report which sources were named and which URLs produced each answer. Our ladder, read from aiknowsus.com/llms.txt on 22 October 2026, is SEO Start at 149 dollars a month in the US and UK or 99 elsewhere, which watches 0 questions on 0 engines, Growth at 299 in the US and UK or 199 elsewhere for 100 questions on all five surfaces weekly, and Scale at 599 and 399 for 300 questions. Growth is the cheapest plan of ours that reads all five surfaces.

This is also the vertical where we are most likely to be the wrong purchase, and we would rather say so. A public agency has staff, a defined list of priority services and no sales target, which means the four measures the searching engine described can be run in house for nothing by one person with a spreadsheet, and that is what we would recommend for the first quarter. If you do get to a procurement, here is where rivals beat us, by name. Profound publishes single sign on, SAML, SOC 2 compliance and up to nine answer engines on its enterprise tier. A word boundary search of our own pricing page, 119,203 bytes read on 22 October 2026, returns zero hits for sso, saml, soc 2, iso, dpa and white label, and for a government buyer those are usually mandatory lines rather than preferences, so expect us to fail that section of the questionnaire. SE Ranking publishes Core at 129 dollars a month billed monthly with 100 prompts tracked daily on seranking.com/subscription.html, read 22 October 2026, against our 299 for 100 questions watched weekly, so on price and cadence alone it is the better line and we did not verify which engines sit behind its prompt count. Our pricing page also contradicts itself, rendering "Businesses covered 1 1 1 1 1 1 10 1" above "More than one business No" in all eight columns, which a county running several agency domains should read before assuming one subscription covers them. Both readings render and we publish both rather than the flattering one.

What this page establishes. One question per engine on one night with the search flag recorded for each, and one fetch per vendor URL on 22 October 2026. Every figure is attached to the page it was read on, and the only price that binds anybody is the one in an awarded contract.

What city and agency staff ask about AI answers

Do we need to buy software

The engine that read 80 pages named no paid product in any of its four steps, and the only directory it pointed at was flagged as a starting point rather than an endorsement. Search Console, Bing Webmaster Tools and GA4 are free and cover everything it asked for.

Which free report shows whether we are being cited

Bing Webmaster Tools and its AI Performance report, which the searching engine named specifically for cited pages, citation trends and grounding queries across Copilot and supported experiences. It is free and verification is a one off task.

What should we measure

The four the engine gave. Citation rate, the share of test questions that cite your official site. Accuracy, whether hours, fees, deadlines and procedures come back right. Coverage, how many priority services appear at all. Outcomes, referral visits and completed tasks. Twenty to fifty questions, the same wording each month, with the cited URLs written down.

How do we stop an assistant giving out a wrong fee

Put the exact figure in text with a publication date, on a crawlable page, and retire the old one rather than leaving both live. One engine's best line was to state exact permit fees instead of calling them low, so there is something unambiguous to quote. Then re test that question monthly.

Why did an answer name a vendor at an address that does not work

Because that answer had not looked anything up. Our log records no live search and zero citations for it, and the hostname it gave for one vendor returned no HTTP status at all when we tried it. Two of its five prices were also wrong against the vendor pages and two of those vendors publish no price. Open the page before any of it reaches a requisition.