---
title: "AI gives wrong information about your business: what to fix"
description: "Imagine a listener asking an assistant about a programme in September 2026 and receiving its 2025 schedule. The cited page was accurate when it was published. N"
canonical_url: "https://www.goodbarber.com/blog/ai-assistants-get-your-business-wrong-fix-the-content-they-cite-a1618/"
lang: en
date: 2026-09-16
last_updated: 2026-09-16
---

# AI gives wrong information about your business: what to fix

[Back](/blog/make-it-r13/)

# AI assistants get your business wrong? Fix the content they cite

Written by [Pierre-Laurent Medori](https://www.goodbarber.com/blog/author/pierre-medori/)  on Wednesday 16 September 2026

## Your current schedule is online, yet an AI assistant tells a listener last year's times and cites one of your own pages. The fix starts in your content: find the page that still presents old information as current, date it and link it to the current answer. With GoodBarber MCP, an agent can search your app's articles for you and apply the corrections you approve.

## Why your own page supports the wrong answer

![](https://cmsphoto.ww-cdn.com/superstatic/41269/art/grande/98047187-68264837.jpg?v=1789542159.3406255)

Imagine a listener asking an assistant about a programme in September 2026 and receiving its 2025 schedule. The cited page was accurate when it was published. Nothing on it says the information has expired, so the assistant treats it as current. The listener misses the programme, or writes to your team to ask which version is right. An old membership offer or a former studio address produces the same confusion.

Carolyn Shelby [raised this issue in Search Engine Journal](https://www.searchenginejournal.com/biggest-ai-search-risk-is-conflicting-information/586904/) on September 7, 2026: contradictory information about a brand can mislead AI search. Three mechanisms explain why your own pages end up in the answer:

- Assistants retrieve pages by wording. A question that uses a programme's former name leads straight to an archive that still contains it.
- They merge several sources into one answer. The dates that separated those sources become hard to see.
- A new page does not retire the old one. The current schedule and the old statement both remain available, and nothing tells the assistant which one applies.

So start with the source the assistant cited. Which sentence on that page no longer describes your service? That sentence needs a date and a route to the current answer.

## A radio schedule that changed in March

Consider this fictional station. In March 2026, it moves a programme to the evening and renames it Evening Edition. The app carries the new schedule. A 2025 archive article, a downloadable press kit and a social profile still describe Morning Edition.

Each copy needs a different treatment. The archive stays, with a dated note linking to the current Programmes page. The press kit and the social profile are still in use, so they get updated. A PDF worth preserving remains online as a clearly labelled archive.

Finding every affected article is the time-consuming part. After a few years of publishing, nobody remembers which articles mention Morning Edition. This is where a connected agent helps.

## Let an agent review and update your GoodBarber articles

If your app runs on GoodBarber, its articles live in the built-in content manager, the CMS. Any change you make there reaches the native app and its web version right away, without a store update. We built [GoodBarber MCP](https://www.goodbarber.com/mcp/) so that an assistant such as Claude or ChatGPT can connect to that CMS and help you maintain the content.

[MCP, the Model Context Protocol](https://www.goodbarber.com/glossary/mcp/), is the standard that lets the assistant read and edit your articles through that connection. You provide the current information. The agent reads the articles, finds the passages that conflict with it and prepares a list of corrections. Ask it to propose first and to write only what you approve.

For the fictional station, give the assistant the current Programmes page and this request:
> Use this Programmes page as the reference. Read my app's CMS articles, including their full text, and find every mention of Morning Edition. Show me the passages that need updating and suggest a note linking to the current schedule. Leave accurate historical references intact. Don't edit anything yet. Tell me which articles you checked and which you couldn't review.

Here is the kind of correction it comes back with.

**Before, in a 2025 archive article:**
> Morning Edition airs every weekday morning.

**Note to add above that passage:**
> Archived schedule for 2025. Since March 2026, Morning Edition has been called Evening Edition and airs in the evening. See the current Programmes page for today's schedule.

The original text stays intact and the note explains what changed. Once you have reviewed the proposal, ask the agent to add the note with a link to the current page, then to read the updated article back so you can check the result. Through GoodBarber MCP, it performs those edits without you opening each article in the back office.

The connection gives the agent access to the CMS content you authorise, and only that. PDFs, store listings and social profiles need their own tools. Ask the agent to list everything it could not check, so those copies stay on your to-do list.

If you have never connected an assistant to your app, [this guide for ChatGPT](https://www.goodbarber.com/blog/use-chatgpt-with-goodbarber-run-your-app-without-code-a1525/) walks through the setup without code.

## Audit your published facts in five steps

Once you have tried the process on one programme, these five steps extend it to the rest of your content.

**1. Start with the questions your audience asks**

List the information your audience needs: schedules, prices, names, addresses.

Use support messages to capture their wording. “Is Morning Edition still on?” belongs in the review even if your team stopped using that name months ago. For each fact, give the agent the current answer and the date it took effect.

[App usage data](https://www.goodbarber.com/blog/mobile-app-analytics-what-to-measure-for-better-decisions-a1602/) tells you which pages people still open. Review those first.

**2. Find every place that repeats each fact**

Ask the agent to search the content it can reach for both the current wording and the older names.

Keep a wider checklist for everything else: website, store listings, PDFs, external profiles, and any page an assistant has cited.

For each match, ask for the article reference, the exact passage and a verdict: current, historical, or needs a correction. Review the proposed changes together before anything is written.

**3. Give each fact a reference page and an owner**

Choose one page that holds the current answer, and name the person responsible for it.

The programme editor owns the schedule, and a new season triggers a review. Wherever possible, other pages link to that reference instead of repeating it. Where the information must appear in full, such as a store listing, add that copy to the owner's update checklist.

**4. Explain the link between old and new names**

Wherever an old name could confuse a reader, add a short explanation: the previous name, its replacement and the change date. If the offer changed as well, say which terms apply to existing members.

These fictional examples show the format:

| Former term            | Current term       | Clarification to add                                                                                                       | Where to review                              |
| ---------------------- | ------------------ | -------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |
| Morning Edition        | Evening Edition    | Since March 2026, Morning Edition has been called Evening Edition and airs in the evening.                                 | Programmes page, archive, press kit          |
| Supporter subscription | Digital membership | Since June 2026, new sign-ups use the Digital membership offer; earlier supporters should check which terms apply to them. | Membership page, help section, store listing |
| Old Studio             | Riverside Studio   | Since July 2026, public recordings take place at Riverside Studio; the address in the 2025 archive is historical.          | Venue page, event archive, social profiles   |

Preserve the original account of a past event. Add its date and a clear route to the current information.

**5. Check for old names regularly**

Keep a short list of previous names for the agent's next review.

A monthly check is a practical starting point. Include the other language versions of your content and any correction left unfinished. If you want the review to run on its own, schedule it in an assistant or automation tool that supports scheduled runs; the MCP connection gives it access to your app.

Repeat the audience questions in an assistant from time to time, and note the answers and the sources it cites. That tells you whether the outdated pages still surface.

Keep your reference articles reachable in the web version of your app as well as in the native app. A public web address is what an assistant can read, cite and hand back to your listeners. [Google's documentation on AI features in Search](https://developers.google.com/search/docs/appearance/ai-features) confirms that the same eligibility rules apply as for classic results.

Make this part of your publishing routine, alongside [long-term app upkeep](https://www.goodbarber.com/blog/will-your-app-still-work-in-three-years-a1573/) and [regular updates](https://www.goodbarber.com/blog/what-quietly-breaks-when-you-don-t-update-your-app-and-why-you-never-notice-on-goodbarber-a1566/). Save the instructions you gave your assistant, so the next schedule change starts with a review you have already defined.

## Where to start this week

Choose one recently renamed programme or changed offer. Follow the [MCP connection guide](https://www.goodbarber.com/mcp-complete-guide/), give your assistant the current information and the reference page, then use the request above.

Review its suggestions, approve the right edits and inspect the result. The next time a listener opens that old schedule, they should find a clear route to the programme they can listen to today.

## Frequently asked questions

**Why does an AI assistant give outdated information about my business?**

Usually because a page, document or profile you still publish presents past information as current. Check the sources the assistant cites for outdated passages. Other causes exist, so start with what you can verify.

**Does publishing a new page fix an incorrect AI answer?**

Not by itself. The new page provides the current information, but older copies remain accessible. Update the ones still in use and add dated notes to the archives worth keeping. Your content becomes unambiguous; when an assistant's answer changes is not something you control.

**How often should I audit these facts?**

A monthly check is a practical starting point. Also review the affected pages whenever you change a schedule, an offer or a location, and check faster-moving information more often.

**Can an AI agent audit and correct my GoodBarber content?**

Yes. Through GoodBarber MCP, a connected agent can read your CMS articles, compare them with the current information you provide and propose corrections. Ask it to apply only the changes you approve, then to read the edited articles back. Its review covers the content it can access.

![Pierre-Laurent Medori](https://blog.goodbarber.com/_public/profile/b5/b5f3704c4117b921f2e8146eddf0b0763b3e1692-default.jpg)

About the author[Pierre-Laurent Medori](https://www.goodbarber.com/blog/author/pierre-medori/)Head of Backend Engineering

I'm Head of Backend Engineering at GoodBarber, where we build the platform that lets anyone create native apps without writing code. My team owns what runs behind the scenes — the APIs, services, and infrastructure that turn a no-code project into real published apps, hosting, push, and a full store. I write about the engineering that keeps "no-code" working at scale, and where AI fits in.

[Read more](https://www.goodbarber.com/blog/author/pierre-medori/)

[![LinkedIn](https://portal.ww-cdn.com/portal_static/svg/base2021/linkedin.3ed8162e2a2b.svg)](https://www.linkedin.com/in/pierre-laurent-medori/)
