Businesses often ask for a switch they can turn on. There is no such switch. ChatGPT may search the web, draw on indexed sources or answer from existing knowledge depending on the question and product experience. The businesses it mentions can also change as sources, wording and context change.
That makes the job evidence-led rather than magical: help systems identify the business, understand what it does, verify the claims elsewhere and find an answer that fits the user’s question.
1. Define the recommendations you actually want
Do not start with ‘rank us in ChatGPT’. Build a prompt map. A Manchester accountant might care about ‘accountant for a growing ecommerce company’, ‘R&D tax adviser in Manchester’ and ‘best accountant for Shopify sellers’. Those are different buying situations and need different evidence.
Record 30 to 50 prompts across problem, service, comparison, location and price intent. Run them without leading language. Log whether your business appears, which competitors appear, which pages are cited and what facts the answer uses. That baseline stops the project becoming guesswork.
2. Make your business easy to identify
Use the same legal name, trading name, address, telephone number, service descriptions and website across your own site and trusted profiles. Connect the organisation to its Companies House record, LinkedIn page, Google Business Profile and relevant professional directories where those profiles genuinely exist.
Your website should include an About page, named people, contact details and Organisation or LocalBusiness structured data that matches the visible page. Structured data describes evidence; it does not replace it.
3. Build one strong page for each buying question
A vague services page gives a search system little to work with. A strong page states who the service is for, the problem solved, what is included, realistic pricing, process, limitations, examples and frequently asked questions.
Write the answer near the top. Use descriptive headings. Include facts that can stand alone when extracted from the page. Avoid padding a simple answer to an arbitrary word count.
4. Earn corroboration outside your own website
A company saying it is the best is advertising. A regulator, recognised directory, customer, partner or reputable publication independently connecting that company with a service is corroboration. Build profiles and mentions where customers in the market already look.
The strongest assets are difficult to copy: verified reviews, named case studies, original research, public tools, professional accreditations and useful contributions to industry publications. Buying hundreds of generic directory links creates volume, not trust.
5. Keep the site accessible and measurable
Check that important pages return a successful status, render useful text without a login, appear in the sitemap and are not blocked from search crawlers. OpenAI documents separate controls for OAI-SearchBot and GPTBot, so decide deliberately what you allow rather than copying a random robots.txt file.
Track referral traffic, assisted enquiries and a stable prompt set. Do not report one favourable screenshot as a ranking. A useful monthly measure is citation coverage: the percentage of tracked prompts where the business is named or cited, segmented by commercial intent.
The takeaway
Start with a defined prompt set, then improve identity, answer pages and independent evidence. Nobody can guarantee a ChatGPT recommendation. A credible provider should show what changed, what was cited and whether qualified enquiries followed.
Questions people ask
Paid advertising and organic recommendations are separate. Buying an advert does not guarantee that an unpaid answer will recommend the advertiser.
There is no fixed timescale. Technical changes may be discovered quickly, while new authority and third-party evidence can take months. Measure a consistent prompt set rather than waiting for one anecdotal answer.
You need useful pages that answer relevant questions. Some belong in an insights section; commercial questions often deserve service, pricing, comparison or case-study pages instead.
Yes. Crawlability, useful content, internal links, reputation and technical quality support both conventional search and AI-assisted discovery.
No. Taprave can audit, implement and track the signals that improve eligibility, but external systems choose their own answers and can change them.