All bookkeeping workflows

Xero's US guide on how to do a bank reconciliation says you compare your business records to the bank statement so every recorded transaction appears on the statement and every bank line appears in the books. It walks through gathering the statement and the ledger, matching deposits and withdrawals, and finishing when the ending balances match. The FAQ gives the formula: bank statement balance plus deposits in transit minus outstanding checks equals the adjusted bank balance, which should match the adjusted book balance after timing differences.

Xero says its automatic bank reconciliation is powered by JAX, which tries to reconcile new statement lines using four methods: Rule (your bank rules), Match (an existing Xero document), Memory (how you reconciled similar items) and Prediction (how other Xero users reconciled similar items). JAX only auto-reconciles when highly confident; otherwise it suggests a match and leaves the final decision to you, and you can challenge or reject its matches.

Anthropic's help article says Claude can execute code to create and work with files directly in a conversation, producing Excel spreadsheets, PowerPoint presentations, Word documents and PDFs that you can download or save to Google Drive. It says code execution and file creation are available on Free, Pro, Max, Team and Enterprise plans, on the web, Claude Desktop and Claude Mobile. Its examples include building financial models in Excel with working formulas and performing advanced analyses on uploaded data, which is what a reconciliation workbook built from bank and ledger exports needs.

A finance educator and former auditor, writing in the Tips & Insights on Finance & AI newsletter, describes dropping 100 invoices and a bank statement into Claude Cowork and running the reconciliation skill from Anthropic's Finance plugin. The post lays out six steps: install the plugin, create Bank_Statement and Invoices subfolders, create a Cowork project, call the skill, ask Claude to add a verification tab, and review only the exceptions. The author stresses that messy inputs still produce poor results.

FloQast describes AI Transaction Matching that matches high-volume transactions across multiple data sources, including bank and credit card reconciliations and 3-way matching. Users design matching rules in natural language and adjust them in an interactive interface without IT support, while automated alerts flag unmatched items and an audit trail records all matching activity.

Booke AI explains that AI bank reconciliation matches bank feed lines to records, suggests categories and points out what does not add up, but finishing the month still requires the statement ending balance to agree and someone to sign off. Booke works inside existing QuickBooks Online or Xero bank feeds, categorizes supported transactions, matches them to records and documents, and leaves low-confidence, unusual or missing-document cases open for review. In QuickBooks Online it prepares the month and you reconcile; in Xero it can also reconcile the line types you configure.

Claude Academy walks through giving Cowork a folder with a checking account statement export and the chart of accounts, then asking it to match transactions to accounts, flag anything that does not line up, and draft journal entries for discrepancies. Claude may ask clarifying questions and build a plan for review, then returns the reconciliation logic, specific discrepancies and journal entries ready for review. Suggested follow-ups include exporting entries as CSV and creating a reconciliation summary for the records.

Ramp's credit card reconciliation guide covers the manual process and then describes how Ramp reconciles Ramp data directly against QuickBooks Online and NetSuite inside the product, so variances surface without exporting to a spreadsheet. Its accounting agent auto-codes transactions across GL, department, class and location using merchant data and historical patterns, assigns a next step to each transaction so the review queue shows only what needs attention, and posts and auto-reverses accrual entries at period end.

Microsoft Learn explains that the Financial Reconciliation agent in the Finance Agent add-in for Excel reconciles two data sets formatted as tables on one or two worksheets. The agent suggests the tables and uses AI to propose reconciliation vectors (mapping and monetary keys) that you can keep, edit or remove before selecting Reconcile data. The resulting report groups transactions as matched, potentially matched or unmatched and includes a generative AI summary, with a Troubleshoot transactions feature for the unmatched items.

Intuit's October 2025 Intuit Connect announcement describes an improved Accounting Agent in QuickBooks with expanded anomaly detection, AI-powered reconciliations, and more accurate automated transaction categorization across classes and dimensions. The release says these agents are rolling out to US-based businesses and accountants in QuickBooks and Intuit Enterprise Suite.

This Workplace AI Institute article argues that bookkeepers should let accounting platforms do bulk bank matching and then hand only the exception list to a general-purpose model such as ChatGPT or Claude. It describes exporting unreconciled items as a CSV and asking the model, using its analysis tool, to find likely duplicates, flag GL coding inconsistent with the payee, and list large items to confirm with the client. The article warns that models can be confidently wrong about specific dollar amounts and that the review is a triage tool, not a posting tool.

Anthropic's prompt engineering overview says to start with three things: a clear definition of the success criteria for your use case, some way to test against those criteria, and a first draft prompt you want to improve. It points to Anthropic's prompting best practices for the techniques themselves, from clarity and examples to XML structuring, role prompting, thinking and prompt chaining. It also notes that not every failing result is best solved by prompt engineering. For adjusting entries, the success criteria are easy to write down: every entry balances, uses only accounts on the chart, and names its support.

Ramp's launch post for its Accounting Agent says it codes every line item, reviews every transaction, and flags anything needing judgment, with routine low-risk spend auto-syncing to the accounting system. At month-end, Ramp accrues transactions that are not ready to sync so spend lands in the right period, reverses the accrual the next month, and reconciles with the ERP, flagging missing, duplicate or mismatched transactions. Corrections can be given once in plain language and the agent learns from them.

Numeric documents an MCP use case where Claude loops through every pending flux task assigned to you, pulls six months of transaction lines per account, groups them by dimensions such as vendor, customer, department, class and location, and drafts conclusion-first variance explanations. Each draft leads with the total variance and primary driver, followed by supporting bullets with dollar breakdowns. Claude drafts and can post the explanation, but does not submit the task; you review and submit.

FloQast's Close AI page describes AI agents that automate repeating close workflows, from pulling data and building entries to tying out and posting, with agents built from a close you already did or from a prebuilt library. It also covers automated amortization, depreciation and subledger reconciliations and continuous GL monitoring that flags anomalous transactions by severity and routes them to the right owner. FloQast says every agent shows its reasoning, an accountant signs off before it runs, and audit evidence is captured automatically.

A Journal of Accountancy article by two University of Akron accounting professors shows how they built an agent in ChatGPT for Excel, with a custom skill, to clean messy general ledger data and reconcile it to the trial balance. The skill makes the agent work from a copy, inspect before acting, present a plan for approval, use native Excel formulas, log every change, and aggregate the cleaned ledger by account to flag non-zero differences against the trial balance. The authors stress human validation, including formula tracing and spot checks.

CPA Practice Advisor reports that Pilot launched Meridian, an AI platform for accounting firms that the company claims completes the month-end close from start to finish and returns review-ready financials. According to Pilot, Meridian closes the books according to each firm's policies and each client's processes, and writes completed output directly into the general ledger. Pilot also says it built and used Meridian internally across more than 8,000 businesses and 187,000 months of books; those are the company's figures, as reported. The article says Meridian has advanced support for QuickBooks Online today and can also work with Xero and NetSuite.

Digits' help center describes Automated Quality Control, which runs checks continuously on the books and surfaces anything that may need review as action items in the Checklist before close. Standard checks are on by default, advanced checks can be enabled, and most checks can be customized with plain English instructions that Digits turns into the logic it runs. Digits' own example checks flag equity balances that change sharply over three months and expense transactions above 1% of annualized total expenses.

Sage Intacct's help documentation explains that Close Assistant, part of Sage Intacct Close Automation, shows which applications are closed and which tasks need attention in the Sage Copilot panel. You open Copilot, select Run Close Assistant, pick the reporting period, and it lists tasks such as reviewing unposted bills and invoices, preparing AP and AR reconciliations, reviewing bank reconciliations, reviewing unposted journal entries and running variance analysis, with links to close each subledger and the general ledger.

Microsoft Learn describes three-way matching as checking the price on a vendor invoice against the purchase order and the quantity on the invoice against the product receipts, with any difference outside a set tolerance flagged as a matching discrepancy. Its example is an invoice billed at 1.10 per unit against a PO price of 1.00 under a 5 percent price tolerance, which fails the match. The Claude Help Center says Claude can run code on uploaded CSV and other data files and create Excel spreadsheets with working formulas that you download from the conversation.

Stampli's blog walks through how its AI assistant, Billy the Bot, extracts invoice data and then codes it. For PO invoices Billy associates the purchase order, pulls GL codes and matches invoice details to the PO, flagging discrepancies; for non-PO invoices it identifies cost center and expense type and flags details it needs clarified, then learns to suggest the right approver over time.

Dext's help centre explains that Auto-publish sends eligible items to connected accounting software after data extraction, based on rules set per supplier, customer, payment method or the whole account. It works alongside supplier rules and Smart split, lets you exclude document types such as supplier statements, applies only to future documents, and Dext AI Assist can also auto-publish using guidance rules that combine conditions like supplier and amount.

BILL's accountant resource article explains how BILL AI extracts invoice details such as vendor name, invoice number, dates, amount, terms and PO number into bills, leaving fields blank when data is not recognized. It also describes duplicate invoice warnings, automatic splitting of multi-invoice documents, a Reminders report that flags potentially missing bills, and a W-9 Agent that emails vendors to collect and enter W-9s.

Zapier's guide explains how invoice automation captures invoices from email or portals, extracts fields, matches against POs, routes approvals and syncs to accounting systems like QuickBooks Online or Xero. It shows Zapier workflows that turn emailed invoices or files dropped in cloud folders into accounting records, log invoices to a spreadsheet and post Slack alerts, and suggests using AI by Zapier to categorize invoices and Zapier Agents to monitor incoming invoices and extract due dates and amounts.

The FBI's Internet Crime Complaint Center (IC3) describes a business email compromise scheme in which a business with a longstanding supplier is asked, by email, phone or fax, to pay an invoice to an alternate, fraudulent account. It advises confirming requests to transfer funds and, when using phone verification, calling previously known numbers rather than numbers in the request. xAI's documentation says Grok's Web Search tool searches the web in real time and browses pages, and its X Search tool searches posts on X and answers with citations. Together they support a pre-payment check: research the vendor's recent public news, then verify any bank change by phone.

Sage Intacct's help documentation explains that when a bill is emailed or uploaded, AP Automation uses Sage Network AI to identify the vendor, extract names, dates, addresses and line items, code GL account, location and department, and create a draft transaction ready to review and post. It also detects and flags duplicates, and its machine learning learns both from collective Sage Network activity and from each company's own corrections.

Anthropic's documentation for the Gmail connector says Claude can search your email and answer questions from what it finds, with citations that link back to the messages it used. It is available on Pro, Max, Team and Enterprise plans, and on Team and Enterprise plans an Owner or Primary Owner enables it before members can connect. Claude accesses only the connected Google account and only email you can already see, retrieves the minimum information needed, and cannot see images embedded in emails. Anthropic's Google Workspace connector help page adds that Gmail access covers attachment metadata, not attachment content.

xAI's announcement of Automations in Grok says an automation can run on an email trigger instead of a schedule. When an incoming email matches your filters on sender, recipient or subject, the automation fires with that email as context and Grok responds to the actual message, then saves the run to its history and reports back by email, app notification, both or neither. The FBI's IC3 recommends company policies that verify any change to existing invoices, bank deposit information and contact information, and contacting requestors by phone before complying with email requests for payments.

Tipalti's finance AI page describes a set of AP agents, including a Bill Approvers Agent that predicts and recommends the right approver, an Invoice Capture Agent, a PO Matching Agent, a Tax Form Scan Agent for W-9 data and a Duplicate Bill Detection Agent. The page says AI learns from past activity to recommend approval paths, works within existing role-based permissions, and that AI outputs may be inaccurate so users should review important outputs.

Vic.ai's case study describes Countsy, an outsourced accounting firm for startups, which Vic.ai says processes up to 3,500 invoices a month across hundreds of clients on NetSuite, Bill.com and Expensify. Countsy adopted Vic.ai to automate invoice ingestion, coding and approval workflows, shifting accountants from manual entry to AI-assisted review, with the AI predicting GL dimensions such as cost account, department and location.

Ramp's help center explains that Bill Pay runs OCR on invoices added by AP forwarding or drag and drop to populate draft bills. On Ramp Plus, Smart OCR references a vendor's historical invoices and saved plain-language custom instructions, and an auto-coding agent predicts accounting fields such as GL category, department and location on each line item based on past bills.

Invoiced's cash application page describes a CashMatch AI algorithm that applies customer payments to the right accounts, invoices and balances, plus remittance advice AI where you upload a document and the system matches payments to invoices while flagging inconsistencies. It also mentions custom matching rules, a virtual lockbox for scanning paper checks, short-pay and over-pay handling, multi-invoice application and a confidence score from 1 to 100 percent per match.

This free interactive lesson has a bookkeeper upload an AR aging report CSV to ChatGPT and ask for a chase list that skips invoices not yet due, holds disputed invoices in a separate list, and sorts oldest bucket first then largest amount. The user checks rows and totals against the file, then asks ChatGPT to draft friendly, firm or final reminder emails, verifies every fact, and sends the emails personally.

Microsoft Support explains that Copilot in Excel can add columns that calculate new values from your existing data, such as durations, and explain how each formula works. A short YouTube tutorial applies this to receivables: Copilot suggests a days-overdue formula, the formula is then edited to count from today when the paid-in-full date is blank and to handle partial payments, and the new column is sorted from largest to smallest to focus on the longest overdue accounts.

Upflow's documentation describes AI collection agents that act on inbound customer emails in the Upflow AR inbox. Each agent has three skills, Replies, Promise to pay and Disputes, and admins set each skill to Off, Suggestions only or Auto-mode, then assign the agent to workflows. Replies must first be managed in Upflow, and the agent works only from customer, invoice and conversation data in Upflow.

Microsoft Learn documents the Business Central Late Payment Prediction extension, which predicts whether posted sales invoices will be paid late so you can adjust payment terms or methods. Once enabled, a Payments Predicted to be Late tile opens Customer Ledger Entries with Late Payment and Prediction Confidence columns, and the model trains on invoice amounts, payment terms, credit memos and aggregated customer payment history.

Chaser's help center explains that its AI assistant, powered by Google Gemini Flash, is a chat tool that can pull receivables data, such as top overdue invoices or highest-status debtors, into tables. It can also create new reminder templates, build chase schedules and assign customers to them, operating on a human-in-the-loop basis where users review and approve templates, schedules and contact moves before anything goes live.

Xero's July 2026 media release from Xerocon London announces new features for JAX, its agentic AI platform. Under Payment Follow Ups, JAX will analyze a customer's payment history and execute tailored collection plans, such as consolidating invoices, timing nudges, or switching to SMS for non-responsive customers. The release also describes Bill Protection, which will inspect bills before payment for unusual amounts, altered bank details or unfamiliar suppliers.

Dext's help centre describes AI Assist as a bookkeeping agent that suggests values for fields like category, tax and description on Costs and Sales items, using document content and past decisions. It explains each suggestion and lets the user accept, dismiss or automate it, with each guidance rule set to manual review or auto-apply. It can also read the body of emails used to submit documents.

Ramp's announcement of receipt automation explains that it can fetch receipts directly from Gmail or Outlook inboxes and receive them automatically from Amazon Business, Uber and Lyft. Its AI parses merchant, amount and date from submitted receipts and combines this with card network data to match receipts to card transactions.

Xero Central explains how to enter details for a document in Hubdoc and publish it to create a new transaction in Xero with a copy of the document attached. It also shows how to enable autosync for a supplier so Hubdoc automatically configures and publishes that supplier's documents to Xero as soon as they are uploaded. Documents detected as credit notes are not auto-published and must be reviewed manually.

This Brocent guide argues that QuickBooks and Xero bank rules should handle recurring transactions deterministically, and that ChatGPT is best used for the residue of unrecognised payees, split transactions and anomaly review. It recommends exporting uncategorised transactions to CSV with the real chart of accounts, asking for a suggested category with a confidence note and flags, and reviewing everything before posting. It stresses never auto-posting AI output, keeping an audit trail and leaving tax judgements to an accountant.

The Firm's guide to ChatGPT prompts for bookkeepers argues that bank feeds and ledger rules already handle matching, so chat assistants are most useful for the writing around the ledger, starting with the email asking a client to explain transactions the bookkeeper cannot code. It recommends grouping similar items, numbering questions, requesting receipts, and following up with a shorter, firmer second chase. The article also says to strip names, tax identifiers and account numbers before pasting, use a firm-approved tier, and never let the assistant decide tax treatment or code from memory.

Anthropic's help center describes Projects as self-contained workspaces with their own chat histories and knowledge bases. Anything uploaded to a project's knowledge base is used across all chats in that project, and project instructions tell Claude how to behave and respond in every chat there. Projects are available on free accounts, which can create up to five. Context is not shared across chats within a project unless it is added to the project knowledge, and on paid plans Claude automatically enables RAG mode when project knowledge approaches the context window limit.

Brex's support article describes Brex Assistant as an AI-powered assistant that can automatically populate and suggest receipts, memos, expense categories and attendees for card expenses. In auto-pilot mode it fills in expense information and assigns spend limits and categories, and it comments on expenses that violate policy to request more information. Managers get natural-language expense search, automated receipt audit and flags on high-risk expenses.

This n8n workflow template monitors Gmail for messages labelled receipts and a Google Drive folder for uploaded receipt files. It uses Anthropic Claude (text or vision) to extract vendor, date, total, currency, category and payment method, flags whether the content is a real receipt, archives attachments to Drive and appends validated expenses to a Google Sheets log with a link to the original receipt.

In this Journal of Accountancy Tech Q&A, a CPA and accounting professor shows an accountant uploading a journal entry log where every expense was coded to Office Supplies. ChatGPT scans the Description column for keywords like laptop and monitor and builds a formula that adds a Reclassified Account column, for example moving those items to Computer Equipment. The author stresses checking the AI's work and never entering client or confidential information into ChatGPT.

Zoho Expense's FAQ explains that Autoscan reads data such as date, amount, currency and merchant from an uploaded receipt and uses it to create an expense. It says Autoscan can extract line-item data and handle receipts in more than 14 languages. Receipts can be uploaded from a computer or cloud storage, dragged and dropped, clipped with browser extensions, or emailed in.

Paychex's guide to running payroll in Paychex Flex describes WISE as an AI engine built into the platform that flags potential issues before submission, such as missing data, unusual amounts and possible compliance gaps. It also automates prep tasks like syncing time data, identifying missing pay rates and generating preprocessing reports. Paychex states that payroll administrators review and approve flags and data before anything is submitted.

This Paychex press release announces WISE as an AI intelligence layer across Paychex Flex, Paycor and SurePayroll that includes agents, assistants and embedded intelligence. Paychex says WISE Agents act within parameters set and controlled by the customer and can schedule shifts and approve timesheets. WISE Assistants provide guidance and task support across chat, voice, email, text and collaboration tools.

ADP's news brief on RUN Powered by ADP describes ADP Assist, which uses generative AI so users can handle tasks like paying employees from the home page search bar in natural language. It also describes Payroll Inspector, which uses GenAI to help catch possible errors in an employee's hours or hourly rates before payroll runs, avoiding re-runs and possible bank fees.

PairWorkflows describes a method for giving ChatGPT payroll totals and general ledger postings for the same period, along with an account mapping and tolerance, so it can draft a reconciliation by cost centre and GL account. It flags items missing from the GL, unmatched GL entries, off-cycle runs and equal-and-opposite variances as candidates to investigate. The page advises minimising personal data, using employee IDs only, and treating the output as an investigation aid rather than a sign-off.

This Bricks guide walks through reconciling a payroll register against bank payout data in Excel with ChatGPT's help. It suggests asking ChatGPT for a reconciliation plan, then for XLOOKUP formulas to pull bank amounts next to net pay, IF formulas to flag differences, and COUNTIF formulas to find payments to people not on the register. It ends with a summary tab showing totals, overall variance and counts of discrepancies.

Deel's blog explains that Deel Payroll runs automated validation as a pre-run step before a payroll package is finalized. AI checks flag duplicates and unusual changes by comparing payroll data against historical baselines, cross-system records and jurisdiction-specific rules. The article stresses that this supports the Payroll Manager's review rather than replacing it, and a person reviews anything flagged before the run closes.

Rippling's launch post for Rippling AI says a user can drop in a messy spreadsheet of spot bonuses and ask it to add them to payroll and convert currencies for review. Rippling AI maps columns to employee records, validates the data, flags missing information, converts currencies and stages the payroll run for approval. The post says changes go through the company's normal approvals and the AI inherits existing user permissions.

Google's help page for Gemini in Google Sheets says Gemini can create tables and formulas, generate data analysis and insights, build charts, and perform actions such as conditional formatting, pivot tables, filters and number formats, each shown on a preview card you apply or undo. It works best with native Google Sheets files, so Google says to save an Excel file as Google Sheets first. Inserted charts go on a new tab with their data and do not change when the original data changes, and the conversation history is lost when you reload or reopen the file. The feature requires an eligible Google Workspace or Google AI plan.

OpenAI's help article on data analysis says ChatGPT can analyze uploaded spreadsheets, including .xls, .xlsx and .csv files, answer questions about the data and create tables or charts. It recommends descriptive column headers in the first row, one row per record, and no empty rows or columns splitting the data. For some tasks ChatGPT writes and runs Python code, and OpenAI says to review the generated code, outputs and assumptions before relying on the result. The article also notes that the number and complexity of formulas in a spreadsheet can affect processing.

Xero Central describes JAX (Just Ask Xero) as a generative AI finance partner that answers questions about a business's financials and about using Xero. With Analytics, JAX can find trends, create cash flow projections, and benchmark the business against similar businesses in its industry. Its feature table lists financial insights such as revenue or profit trends, past and projected cash flow position, and industry comparisons. JAX is in beta, needs the admin or standard user role, and its features depend on the Xero pricing plan.

This TechNews.Tax newsletter post describes how a tax firm reviews Profit and Loss statements from clients who do their own books, using an AI prompt that compares the P&L to typical expense categories and percentages for the client's NAICS code. The prompt asks for a NAICS summary, a category and percentage comparison, a judgment on whether expenses are in line (within a 10% variance), and a risk explanation for outliers. The author uses the OpenAI API but notes the same process works in ChatGPT, Gemini, Copilot or similar tools.

In this Mostly Metrics tutorial, the author records building a 13-week rolling cash forecast for a fictional software company from eight messy data tabs, including an invoice export with inconsistent terms, collections history, AP, a payroll calendar and one-off items. Claude first works out each customer's actual days-to-pay from collections history rather than nominal terms, then builds a forecast with beginning cash, collections, disbursements, net change and ending cash chained week to week. The model includes a covenant line with an automatic OK/BREACH flag.

A Journal of Accountancy technology Q&A shows how ChatGPT-4o can create Excel spreadsheets from scratch and analyze uploaded files. In the financial analysis example, the author pasted raw financial data and ChatGPT returned a workbook with tabs for Raw Financials, Horizontal Analysis, Trend Analysis and Ratio Analysis. The author advises being specific in the prompt if you want working formulas rather than values.

CPA Practice Advisor walks through pasting an anonymized landscaping company's income statement into ChatGPT and Gemini Advanced and asking how financially healthy the business is and what to recommend. Both tools identified the net loss, flagged salaries as a large share of expenses, and suggested cost and revenue actions. The author stresses anonymizing data, getting client permission, having an expert check every answer, and reviewing terms of service.

Fathom's help centre explains how the Commentary Writer, an AI tool inside Fathom Pro reports, drafts data-based commentary from up to six selected report components. Users add persistent Business Context (goals, strategy, market conditions) and per-period Report Context (one-off events, audience), choose a Simple or Technical style and length, then prompt and refine before adding the text to the report. Figures in the generated commentary can be hovered over to check where they came from.

Microsoft Support explains that Copilot in Excel lets users ask questions in their own words to surface summaries, trends and outliers, returning results as charts, PivotTables or summaries. It can also add formula columns and rows (for example profit margin or percent of total), suggest single-cell formulas, build lookups such as XLOOKUP, and explain existing formulas. Microsoft advises being specific about which columns to analyze and to review and verify anything Copilot creates.

Spotlight Reporting's help article describes AI options on the Executive Summary page and in Text Panels that can generate commentary on a report for up to five topics, or rewrite existing commentary to be shorter, longer, or more professional. The article notes that AI-generated text reflects the values at the time it was created and does not update automatically when data changes, unlike tags and auto-text. It also recommends checking generated text before completing the report because the AI can occasionally make mistakes.

Anthropic's documentation describes Claude for Excel as an add-in for Pro, Max, Team and Enterprise plans that brings Claude into Excel. It can answer questions about an open workbook with cell-level citations, adjust assumptions while keeping formula relationships intact, identify errors and their root causes, and work across multi-tab workbooks. It can also sort, filter, edit pivot tables and add data validation dropdowns. The documentation warns that files from external sources can contain hidden instructions that try to manipulate the add-in, and says to start with a trusted copy before asking Claude to edit widely.

Google's Docs Editors Help explains that Gemini in Google Sheets can create tables and formulas, generate data analysis and insights, build charts, and perform actions such as creating pivot tables, applying conditional formatting, sorting, filtering and find and replace. Users open the Ask Gemini side panel, describe the formula they need with references to their sheet, and insert or retry the suggestion. When a formula shows an error, users can hover over the cell and click Fix so Gemini reviews the formula and analyzes the issue.

Google's Business Profile Help says to pick a primary category that best describes the business, choosing the most specific option on its list, and notes you cannot create your own category. You can add up to 9 more categories, but Google says not to use categories as keywords or to describe features. Changing a category may trigger re-verification. Attributes describe the business, some factual ones can be edited, and subjective ones come from Google users.

Google Search Central documents LocalBusiness structured data, which can tell Google a business's name, address, hours and other details, and lists required and recommended properties. Google's Business Profile Help says local results are based mainly on relevance, distance and prominence, and suggests keeping business information complete and accurate. Google announced in August 2023 that FAQ rich results would show only for well-known, authoritative government and health sites, so FAQs on a firm's page help readers rather than earning a rich result.

These review reply templates follow Google's Business Profile Help: a verified business can reply to reviews from its profile. Google checks each reply against its content policies before posting, which usually takes up to 10 minutes but can take up to 30 days. An approved reply is public, shows as the business rather than a personal name, and the reviewer is notified and can still change the review.

Google's Business Profile Help describes three post types: Updates, Offers (title, dates and time required, with a View Offer button) and Events (title, start and end dates). Posts can be scheduled or set to repeat, and Google reviews each post against its content policy before it goes live. Google says posts with a phone number in the text may be rejected, and that posts older than 6 months are archived unless a date range is set.

Google's Business Profile Help says reviews are useful to potential customers when they are honest and objective, and that a mix of positive and negative reviews is more trustworthy. It says to value all reviews, reply to them, and remind customers to leave reviews, and that business owners should not offer incentives. Google's Maps content policy prohibits fake engagement, which rules out writing or buying reviews for your own or a competitor's profile.

Google's Business Profile Help says the description should cover what you offer, what makes you special and how long you have been in business, must stay within 750 characters, and must not include URLs or HTML. Google's guidelines add that descriptions should not focus on promotions, prices or sales, and should not show links or gimmicky characters. Google also offers AI-written descriptions in some regions and says to review them for accuracy.

xAI's developer overview describes the Responses API for generating text, multi-turn chat and function calling, alongside its voice, image and video APIs. The quick start sends a request to https://api.x.ai/v1 with an xAI API key, using either the OpenAI SDK pointed at that base URL or a curl call, and the docs also cover web search, structured outputs and the Batch API. This workflow uses the API, not the Grok app: a small script or internal tool sends the client's question with the books data you choose to include, and you review the draft answer before anyone sends it.

OCTA's resource page describes AI chart of accounts cleanup that reviews a client's account list against actual posting activity to find duplicates, near-duplicates, unused accounts, wrong account types, misparented sub-accounts and naming inconsistencies. In OCTA Flow the agent connects to QuickBooks, Xero, Sage or Zoho, proposes merges, reclasses, deactivations and renames ranked by severity, and the bookkeeper applies approved changes in the ledger. The page notes that merges can move history and affect prior-period reports, so those are flagged for close review.

xAI's Grok 3 launch post introduced DeepSearch as its first agent, built to synthesize key information, reason about conflicting facts and opinions, and distill clarity from complexity, including the latest real-time news. xAI says Grok 3 models were equipped with code interpreters and internet access so they can query for missing context and adjust their approach, and that DeepSearch ends with a concise, comprehensive report. This workflow applies that to a bookkeeping decision: a research brief comparing two apps on pricing, integrations, data export and recent complaints, with every claim tied to a link you can check.

Karbon's February 12, 2025 release notes announce two open beta Karbon AI features for accounting firms. Karbon says quick replies offer three suggested responses to common incoming emails, and that AI summaries cover the 10 most recent conversations on each work item. The summaries are meant to help users get up to speed before client meetings without sifting through emails and notes. Karbon positions the conversation summaries as especially useful for owners, managers and partners overseeing many client relationships.

Financial Cents' guide to ChatGPT for accounting firms lists writing SOPs, checklists and workflow documentation as a practical use case, turning how a task is already done into a repeatable procedure another team member can follow. Its example approach asks the assistant to use only the steps described and to ask about gaps rather than invent steps. The article also notes that ChatGPT is a general-purpose chatbot whose answers can be generic compared with AI built into practice management software.

xAI's cookbook example builds a fact-checker with the Grok API. You paste a link to a post on X, and Grok reads it with X Search, images and video included, and splits it into up to four claims. It then checks each claim on the web and on X in its own request and writes a note in which every sentence cites a source the searches actually returned. Each claim gets a verdict of supported, missing context or not enough evidence, and any source Grok lists that the searches did not return is dropped, so a verdict stands only when a kept source backs it. The source-checking safeguards described here are part of the cookbook code, not the Grok app.

Avalara's blog introduces Avi for Tax Research, a conversational AI agent embedded in Avalara Tax Research and trained on Avalara's compliance content. Users ask product or service taxability, exemption and rate questions and receive explanations with links to supporting citations. The post's example shows the assistant retaining context so a user can ask whether the same transaction is taxable in a second state without restating the scenario.

xAI's developer documentation for the X Search tool says Grok can run keyword searches, semantic searches, user searches and thread fetches on X through the API, then answer with citations. A request can set allowed_x_handles to consider posts only from up to 20 named accounts, plus from_date and to_date, and those limits apply to every keyword and semantic search the model writes, whatever query it chooses. The IRS social media page lists its X accounts, including @IRSnews, @IRStaxpros and @IRSsmallbiz, and says its posts include tax-related announcements for individuals, businesses and tax professionals, plus the latest scam alerts.

xAI's announcement of Automations in Grok describes jobs Grok runs on its own. You write the instructions once, like a chat message, then choose a one-time, daily, weekday, weekly, monthly or yearly schedule in your timezone. Every run is a fresh request with current data, saved to run history, and Grok reports back by email, app notification, both or neither. The IRS says it lists the latest disaster relief, with the eligible localities, on its Tax relief in disaster situations and Around the nation pages, and that administrative relief follows an emergency declaration or a major disaster declaration offering FEMA Individual Assistance.