Clients have started asking the question: "What about ChatGPT? Can people find us there?" For an agency or freelancer, it is both an opportunity and a trap. The opportunity is a new, genuinely useful service. The trap is promising outcomes nobody can control, or delivering a vague report that does not change anything.
By the iSuggest.ai Team ยท Updated for 2026
This post is a practical process for auditing a client's site for AI search readiness: how to scope it, what to check, how to report it, how to prioritise and what to promise. It is written so that you can reuse it across clients and hand parts of it to junior colleagues. We are careful about claims throughout, because the credibility of your service depends on not overselling it. For background on the discipline itself, see what GEO is.
Why offer this audit at all
Clients are not wrong to ask. Industry compilations report that roughly 68% of US Google queries are zero-click (Digital Applied), that AI Overviews appear on about 48% of tracked queries (SEO Sherpa) and that AI referral traffic grew about 16x between 2024 and 2026. These are third-party compilations with their own methods, so present them as reported figures and not as settled facts. The honest message to a client is: this channel is growing from a smaller base, it is not a replacement for search, and there are concrete, inexpensive things that make a site easier for assistants to use.
The audit is also a natural entry point. It is bounded, it produces a clear deliverable, and it often uncovers problems, such as blocked crawlers, that benefit the client regardless of AI.
Step 1: Scope the engagement
Agree what you are auditing before you start. A short scoping conversation prevents most disappointments.
- Goal. What does the client want: more qualified enquiries, brand recognition, protection of existing traffic? Name one primary goal.
- Pages. Choose the homepage, the key service or product pages, the contact or location pages and a sample of content pages. Fifteen to thirty pages is plenty for a small site.
- Questions. Ask the client for the ten to twenty questions customers ask. These become your test prompts later.
- Competitors. Ask for three or four names the client considers peers, for comparison in assistant answers.
- Access. Confirm who can change robots.txt, the firewall, the theme and the markup, so recommendations reach someone who can act.
- Expectations. State at the outset that the audit improves readiness and cannot guarantee citations.
Step 2: Check access
If crawlers cannot get in, nothing else counts, so start here.
- Read robots.txt. Note any rules affecting AI user agents and whether they look intentional. One report from LLM Pulse found that over 35% of the top 1,000 sites block at least one major AI crawler, so expect to find some blocks. See robots.txt syntax mistakes.
- Ask about firewalls, content delivery networks and bot-protection plugins that may block automated traffic regardless of robots.txt.
- Confirm important pages are not behind logins, paywalls or interstitials.
- Record response codes, redirect chains and any error pages on key URLs.
Remember that blocking can be a deliberate client decision, for example to protect licensed content. Report it and let the client decide; do not change it silently. Background reading: the rise of AI crawlers and how to check if GPTBot can crawl your website.
Step 3: Check rendering
Many AI crawlers read raw HTML and do not run scripts. For each key page, compare what a browser shows with what a plain fetch returns.
- Is the main text, heading structure, pricing and contact information in the raw response?
- Do navigation and internal links exist as real links in the HTML?
- Are important facts locked inside images, PDFs or script-built widgets?
Single-page applications and heavily scripted themes are the usual culprits. See what client-side rendering hides from AI crawlers.
Step 4: Check identity and structured data
An assistant needs to know what the client is, in one consistent form.
- Does the homepage state plainly who the client is, what it offers, for whom and where?
- Is Organization or LocalBusiness markup present, valid and consistent with visible content?
- Are name, address, phone and description identical across the site and across the main external profiles?
- Do key pages have appropriate Article, Product, Service or FAQ markup, matching what is on the page?
- Are authors and contact routes visible?
See entity clarity and structured data for AI search engines. Flag any markup that claims things the page does not show, because it is worse than none.
Step 5: Check the content itself
Review key pages for how usable they are as a source.
- Does each page answer a real question near the top?
- Are facts specific: prices or ranges, timelines, coverage, policies, specifications?
- Are passages self-contained enough to be quoted alone?
- Are there dates, and are they current? Is anything demonstrably out of date?
- Is there a genuine FAQ where customers ask follow-ups?
Our guides on writing content AI models want to quote and FAQ sections for AI search give you language to reuse in recommendations.
Step 6: Check technical health
Speed and URL hygiene still count, because agents work to time limits and avoid messy sources.
- Load speed on key pages. See why page speed still matters for AI search.
- Canonical tags, duplicate URL variants and parameter handling. See canonical tags and URL variants.
- Sitemap present, accurate and referenced.
- Mobile usability and a clear heading hierarchy.
Step 7: Check reputation and mentions
What others say about the client matters, and it is the one area an audit of the site alone cannot see.
- Search the client's name and list independent descriptions: reviews, press, directories, partner pages, community threads.
- Note consistency of the description across them.
- Compare against the named competitors.
The evidence here is suggestive. One analysis reports brand mentions correlate with AI search visibility about three times more strongly than backlinks, 0.664 versus 0.218 (Media Officers), and Meltwater reports earned and news media at 37.6% of citations versus 0.2% for press releases across more than 8 million citations and eight models (Meltwater). Present them as one study each, with the limits stated. See brand mentions versus backlinks.
Step 8: Test real answers
Take the client's customer questions and ask several assistants. Record, for each question and each assistant: whether the client is named, what is said, which sources are cited and which competitors appear.
- Use the same wording each time, and record the date.
- Repeat each prompt more than once, because answers vary between runs.
- Treat the results as a baseline snapshot, not a score. Do not average them into a single made-up number.
- Note inaccurate statements about the client, which are as valuable a finding as absence.
Make the limits plain in the report: assistants differ, answers change, and a single run proves little. Patterns across weeks are what matter. For analytics, see how to measure AI search traffic in Google Analytics.
Step 9: Report so the client can act
A long report that nobody reads helps no one. Aim for something a busy owner can use in ten minutes, with detail in an appendix.
A suggested structure
- Summary. Three to five sentences: where the site stands and the top three actions.
- Findings by area. Access, rendering, identity, content, technical, reputation. For each: what you checked, what you found and why it matters, in plain language.
- Answer test results. The baseline table, with caveats.
- Prioritised actions. Ordered list with owner, effort and expected benefit.
- What we did not check. Be explicit about scope limits.
- Next review date. Audits are snapshots; agree when to repeat. See how often to re-audit your site.
Tone
Describe problems without drama and avoid jargon. "The site's firewall rejects automated requests, so assistants may not be able to read your pages" is better than a list of acronyms.
Step 10: Prioritise sensibly
Order matters because each layer depends on the one before it.
- Access issues. Fix first; they block everything else.
- Rendering issues. If the text is not in the HTML, content work is wasted.
- Identity and consistency. Cheap and high value.
- Content improvements. Start with the pages closest to revenue.
- Reputation work. Slow and ongoing; begin early but expect results over quarters.
Within each layer, rank by effort against benefit, and group quick wins into a first sprint so the client sees progress. A 90-day structure helps; see the first 90 days plan, and the AI SEO checklist for a longer list of checks.
Step 11: Set honest expectations
This is where reputations are made or lost. Say clearly:
- The audit and fixes improve readiness. They do not guarantee that any assistant will mention the client.
- Assistants change their behaviour often, so results can move without any change on the client's side.
- Reputation work takes months.
- Classic SEO remains important; GEO is built on the same foundation. See SEO versus GEO versus AEO and ranking is not enough.
Avoid promising rankings, citation counts or traffic numbers, and avoid guarantees offered by tools or competitors. A client who understands the limits will trust you more when things do improve.
A worked example
Imagine an agency auditing a small independent veterinary clinic. This is an illustration, not a case study.
Scoping settles on the homepage, five service pages, the contact page and four blog posts. The client supplies fifteen questions, such as "Do you offer emergency appointments?" and "How much is a dental clean for a dog?". The audit finds that a security plugin challenges automated requests, the price list is an image, the homepage never lists the neighbourhoods served, and the clinic appears under two slightly different names on its two main directory listings. Asked the fifteen questions, assistants name the clinic in a few answers, describe one service incorrectly and often recommend competitors with clearer pages.
The report leads with a summary and three actions: allow the crawlers the clinic wants, turn the price image into a text table, and standardise the name and add a service-area sentence. Further actions follow in priority order, including a short FAQ and a plan to request reviews from satisfied clients. The agency states that none of this can guarantee a mention, schedules a repeat answer test in three months, and agrees to compare against the baseline.
Common mistakes agencies make
Selling guarantees
Never promise citations or placement. It is the fastest way to lose a client.
Delivering a tool printout
A raw automated report without interpretation or priorities is not a service. Add judgement.
Skipping the answer test
Clients care about what assistants actually say. Showing real answers makes the audit concrete.
Ignoring the client's own rules
Some clients intentionally block AI crawlers. Respect that and report, do not override.
One-off thinking
A single audit ages quickly. Offer a light recurring review instead.
Speed up the technical part
Much of steps 2 to 6 can be done quickly with the free iSuggest.ai audit, which checks crawler access, rendering, structured data, headings and identity signals in one pass. You can read what an audit measures for the full list. Audits are free and not metered, so you can run one on every page in scope, re-run after fixes and use the before and after as evidence in your report. You still supply the judgement, the answer test and the reputation review.
For clients who want a consistent, crawlable description of their business, you can publish a factual snapshot to the iSuggest.ai AI directory, with pages for ChatGPT, Claude and Gemini. Publishing spends credits, which are non-refundable, and improves findability without guaranteeing a citation; see how credits work. See how it works, try an audit on a client homepage, then sign in at your account page when you are ready.