ChatGPT SEO: the honest guide for local businesses

Most agencies selling 'ChatGPT SEO' are selling regular local SEO with an AI rebrand. The good news is that there is a real methodology for influencing what ChatGPT (and Perplexity, Claude, Gemini, Bing Copilot) say about your business. The bad news is that it looks nothing like what gets you ranked on Google.

This guide lays out the actual high-leverage moves, in priority order, with no padding. By the end you'll know exactly where to spend your next 5 hours of effort.

How LLMs actually find businesses

There are two distinct mechanisms. The first is training data: every major LLM is trained on a snapshot of the web (and Reddit, Wikipedia, Wikidata, podcast transcripts, YouTube auto-captions). If your business name appears in those sources, alongside your service and city, the model 'knows' you.

The second mechanism is live retrieval. ChatGPT Browse, SearchGPT, and Bing Copilot all use Bing's live search index. Perplexity has its own index. Google Gemini uses Google's. When you ask one of these models for a recommendation in real time, they search, read the top results, and synthesise an answer.

Both mechanisms reward different signals. Training data rewards brand mentions across high-trust sources. Live retrieval rewards being indexed and being ranked well on the underlying search engine.

The high-leverage moves, in order

Here's the actual list, ranked by impact-per-hour for a typical local service business.

• 1. Claim Bing Places. ChatGPT Browse uses Bing's index. This is the single most-ignored action in 'AI SEO'. 30 minutes at bingplaces.com.
• 2. Verify in Bing Webmaster Tools. Submit your sitemap. If Bing doesn't crawl you, ChatGPT can't see you. 20 minutes.
• 3. Create a Wikidata entry. Wikidata is a structured-data sibling of Wikipedia, used by Google Knowledge Panel, Siri, Alexa, and every LLM trained after the entry date. 60 minutes at wikidata.org/wiki/Special:NewItem.
• 4. Add the business to OpenStreetMap. OSM data is used by Apple Maps, Mapbox, and bulk LLM training corpora. Free, permanent. 30 minutes.
• 5. Submit your key pages to the Wayback Machine. Wayback snapshots get re-crawled by training data builders. A snapshot today = a permanent record. 15 minutes.
• 6. Cross-link via schema sameAs. In your LocalBusiness JSON-LD, add a sameAs[] array containing Google Maps URL, Wikidata Q-number, every social profile, Crunchbase, Trustpilot, BingPlaces, AppleMaps. This wires your entity graph together. 20 minutes.
• 7. Build genuine Reddit presence. OpenAI signed a data-access partnership with Reddit in 2024, and Reddit content appears widely in LLM training corpora via Common Crawl. Reddit threads are also a heavily-cited source for Perplexity and Bing Copilot answers. Build karma for 4+ weeks before mentioning your business. Be honest, disclose, never spam.
• 8. Pitch local press + trade publications. One mention in The Guardian, a regional newspaper, or your trade body's magazine outranks 50 directory listings for LLM authority. HARO/Qwoted/SourceBottle are the easiest entry points.
• 9. Get on 3 podcasts in 90 days. Podcast transcripts are LLM training gold. A 45-minute guest spot = 6,000+ words of training data with your name mentioned 10+ times.
• 10. Run a review programme on Google + Trustpilot. Review velocity feeds both live retrieval (Bing/Perplexity cite Trustpilot heavily) and training data.

What NOT to do

These are the moves that look like 'AI SEO' but actively backfire:

• Stuff your homepage with 'ChatGPT' or 'AI' keywords. LLMs detect keyword stuffing and devalue.
• Buy backlinks from PBNs or low-quality directories. Bing penalises hard and ChatGPT Browse skips low-trust domains.
• Spam Reddit with business mentions. Permanent shadow-ban.
• Auto-generate FAQ pages with AI tools. LLMs detect their own output; the pages get devalued.
• Cloak your address with PO Boxes or virtual offices. Bing detects NAP mismatches and drops you from local results entirely.

Why this works

LLMs are pattern-matching machines trained on a web they've already seen. When your brand appears consistently next to your service and city across high-trust sources (news, Wikipedia, Reddit, trade publications, official directories), the model learns the association. When asked 'best plumber in Bolton', it returns the associations it has the most evidence for.

Live retrieval is simpler: be in Bing's top 5 results for the underlying query, and you'll often appear in ChatGPT Browse's answer.

Both mechanisms reward the same underlying thing: real presence on real sources. The shortcuts don't work because the LLMs are good at detecting shortcuts.

How to measure progress

Don't trust agency dashboards. Test in the actual models. Open a fresh incognito window. Run these 10 prompts in ChatGPT, Perplexity, Claude, Gemini, and Bing Copilot. Log whether your business is named, the position, the sentiment.

• Best [service] in [city].
• I'm looking for [service] near [city]. Who would you recommend?
• Tell me about [your business name] in [city].
• What do you know about [your business]? Are they reputable?
• How does [your business] compare to other [service] in [city]?
• I need to book [service] this week in [city]. What are my options?
• What's the most trusted [service] provider in [city]?
• Where can I find [your specialty] in [city]?
• Are there any [service] in [city] I should avoid?
• I just moved to [city]. Which local businesses should I know about?

Run them weekly. Track Week 0, Week 4, Week 8, Week 12. By Week 4 you should be named in 2-3 of the 10 prompts. By Week 12, 7+ of 10. That's the curve.

If you'd rather not manage this yourself, that's exactly what Asked exists for. The free audit shows you where you stand today. The £99 checklist gives you the entire 23-sheet implementation playbook, pre-filled for your business. The £499 service has us run the playbook for you in 14 days, with a money-back guarantee tied to score movement.