AI and Local Search: How Practitioners Use AI and What They Expect from SGE
Generative search arrived in search engines quickly in 2023, and local SEO teams had to decide which AI uses were safe to adopt and which to leave alone.
This guide collects the practical views of five working practitioners, then turns them into a decision checklist you can apply to your own workflow. The views were recorded in mid-2023, when Google's Search Generative Experience (SGE) was still an opt-in experiment, so treat the SGE commentary as a dated snapshot rather than a forecast of current behaviour.
How practitioners were using AI in mid-2023
The five contributors converged on a similar pattern: AI assistants were useful for research, structuring, and repetitive analysis, but not for publishing customer-facing content. Ben Fisher described AI as an assistant for research, technical tasks, and idea generation for writers, with prompts collected into a reusable library so the team could spend more time on strategy. Dayna Lucio used AI mainly to test platforms and to streamline tasks such as summarising data, mapping cities to ZIP codes for local tracking, and building regex filters for Google Search Console; she also used it to brainstorm FAQ questions and metadata tests. Amanda Jordan used AI to generate ideas and to think about data in new ways, and argued that understanding how generative AI works is itself a useful SEO skill. Jess Peck's team applied machine learning across analysis, clustering, forecasting, text analysis, and some content generation, combining techniques such as LDA, keyword extraction, and text summarisation with Google APIs and TensorFlow, and chunking LLM outputs before checking them against the input with other methods. Andy Simpson's team used ChatGPT for content outlines and ideas, and to expand and suggest additions to local business schema, verifying the output before use.
A useful way to read these accounts is as a risk ladder. Low-risk uses are internal: summarising data, drafting outlines, generating regex, clustering keywords, and proposing schema snippets that a human then checks. Higher-risk uses touch the customer directly, such as publishing AI-written blog content or letting an assistant answer local queries without review. None of the five contributors recommended the higher-risk end.
A worked example: schema drafting with verification
One concrete example from the captured evidence shows the pattern in practice. Andy Simpson's team used ChatGPT to expand and suggest additions to local business schema, and specifically found it effective for writing FAQ and HowTo schema code. The important part is what happened next: the team verified what was produced before using it. That two-step pattern — draft with the assistant, verify with a human — is the same discipline the other contributors described for prompts, clustering, and content outlines. If you copy the pattern, keep the verification step explicit: check the generated markup against the schema you actually intend to publish, and do not let the assistant's confidence substitute for a check.
What the practitioners expected from SGE
Expectations were mixed and, in several cases, cautious. Ben Fisher saw a glimmer of hope because the 3-pack appeared to match regular search results, but expected new winners and losers as Google learned, and worried that small and medium businesses would lose the real estate they were used to. Dayna Lucio pointed to Google's restaurant comparison layout as evidence that SGE can work at a local level, and argued that the fundamentals — complete Google Business Profile information, on-page content, and attention to reviews — remain the levers, with query testing needed once more people have access. Amanda Jordan hoped SGE would push sites away from generic content toward unique, personalised, usable material, and stressed understanding customers and the customer experience. Jess Peck was openly ambivalent, doubting that these experiences usually improve sites and arguing that sites relying on visitor metrics must focus on what a machine cannot replace, such as human, comprehensive reviews and small hobbyist sites with a human touch. Andy Simpson voted 'Meh' in a public poll but liked the direction, treated it as version 1.0, and was not panicking about whether a client appeared in SGE results while access was still limited.
Read together, the group did not predict a single outcome. The common thread is that SGE raised the value of distinctive, human-verified information and lowered the value of generic content, while the commercial effect on small businesses remained uncertain.
A decision checklist for your own workflow
Use the following checklist to decide which AI uses to adopt first. It is derived from the practitioners' accounts, not from a controlled test.
- Classify the task. Is the output internal (research, summaries, regex, clustering, outlines) or customer-facing (published content, live answers)? Start with internal tasks.
- Name the verification step. For every AI-assisted output, write down who checks it and against what. The schema example shows the pattern: draft, then verify before publishing.
- Keep prompts reusable. Fisher's prompt vault is a low-cost way to make the assistant's output more consistent across the team.
- Separate 'assistant' from 'author'. None of the contributors treated AI as a replacement for content creation; keep a human author responsible for anything published.
- Protect the local fundamentals. Complete Google Business Profile information, on-page content, and reviews were named as the levers that SGE continues to reward.
- Test queries, not just rankings. Lucio's advice to test different local query types once access widens is a practical way to see how answers and layouts change.
- Watch small-business exposure. Fisher's concern about lost real estate is a reminder to monitor how much visible space remains for SMB clients as SGE evolves.
Two different failure modes to keep apart
The accounts point to two distinct risks that are easy to conflate. The first is an output-quality failure: the assistant produces something plausible but wrong, such as schema that does not match the intended markup or a summary that misstates the data. The fix is verification, which is why the schema example and the chunk-and-check approach both include a human or secondary-method step. The second is a visibility failure: even correct, well-verified content may lose prominence if SGE changes how much space local results occupy. The fix there is not better prompting; it is monitoring query layouts and investing in what a machine cannot easily replace, such as human reviews and distinctive local information. Treating the second as if it were the first leads teams to over-polish content while missing the structural shift.
| Practitioner | Role | Current AI use | SGE view |
|---|---|---|---|
| Ben Fisher | Founder and VP of Marketing, Steady Demand | Research, technical tasks, idea generation for writers; prompt vault; no forward-facing AI content | Some hope from 3-pack parity, but expects new winners and losers and pressure on SMBs |
| Dayna Lucio | Strategist, SEO, Amsive Digital | Platform testing, data summarising, city-to-ZIP mapping, GSC regex, FAQ and metadata brainstorming | Interested; local restaurant comparison shows local potential; fundamentals and query testing matter |
| Amanda Jordan | Director of Digital Strategy, RicketyRoo | Idea generation and new ways to think about data; understanding generative AI capabilities | Hopes SGE ends generic content and rewards unique, personalised, usable content |
| Jess Peck | Machine Learning Engineer, Local SEO Guide | ML for analysis, clustering, forecasting, text analysis, some content generation; LDA, keyword extraction, summarisation, Google APIs, TensorFlow, LLMs with chunking and checks | Ambivalent; doubts these experiences improve sites; favours unique, human, hard-to-replace content |
| Andy Simpson | Senior SEO Specialist, Digital Law Marketing | ChatGPT for outlines and ideas; expanding local business schema; FAQ and HowTo schema code with verification | Voted 'Meh' but likes it; treats it as version 1.0 and is not panicking while access is limited |
Recorded in mid-2023 when SGE was an opt-in experiment; views are a dated snapshot and do not describe current search behaviour.
Frequently asked questions
Should local SEO teams use AI to write published content?
The practitioners in this snapshot did not recommend it. Fisher kept AI away from forward-facing blog content, Simpson's team used ChatGPT for outlines and schema rather than final copy, and the closing summary stated that generative AI should not replace content creation. The consistent pattern is to use AI for drafting and structure, then have a human author and verify the result.
Does SGE change the local 3-pack?
In the mid-2023 snapshot, Fisher observed that the 3-pack appeared to match regular search results, which he read as a glimmer of hope. He still expected new winners and losers as Google learned. Because SGE was an opt-in experiment at the time, this is a dated observation rather than a description of current behaviour.
How did one team use ChatGPT for local business schema?
Simpson's team used ChatGPT to expand and suggest additions to local business schema, and found it effective for writing FAQ and HowTo schema code. They verified the output before using it. The reusable pattern is draft with the assistant, then check the generated markup against the schema you intend to publish.
What did the practitioners fear for small local businesses?
Fisher worried that small and medium businesses would suffer as the real estate they were used to was replaced by new features. Peck argued that sites relying on visitor metrics must focus on what a machine cannot replace, such as human, comprehensive reviews and small hobbyist sites with a human touch. Neither view was a prediction of a specific outcome.